View Profile
View Forum Posts
Private Message
[NEW] FFmpegForge – A Powerful FFmpeg GUI for Video, Audio and Image Conversion
Hi everyone!
I'd like to introduce FFmpegForge, a Windows graphical user interface for FFmpeg that I've been developing.
FFmpegForge is designed to make FFmpeg's extensive multimedia conversion capabilities accessible through a modern, user-friendly interface, while giving users the freedom to choose their preferred formats and encoders.
Unlike converters that rely on a predefined selection of output formats, FFmpegForge dynamically detects the formats, codecs, and encoders available in FFmpeg.
This is one of the main ideas behind the project: instead of restricting users to a small number of commonly used formats, FFmpegForge aims to expose the extensive capabilities of FFmpeg directly through its graphical interface.
Main Features
Dynamic FFmpeg Format and Encoder Detection
FFmpegForge automatically retrieves the available formats, codecs, and encoders from FFmpeg.
- No fixed, hardcoded list of supported formats.
- Access to common and uncommon multimedia formats.
- Automatic discovery of capabilities provided by the FFmpeg build.
- Support for newly available formats and encoders as FFmpeg evolves.
The available capabilities depend on the FFmpeg build being used.
Video Conversion
Convert video files using the formats and video encoders available in FFmpeg.
Users can select their preferred output format and video encoder instead of being restricted to a small collection of predefined conversion profiles.
Audio Conversion
FFmpegForge also provides access to FFmpeg's audio conversion capabilities.
Users can choose from the available audio formats and encoders, including less common options supported by FFmpeg.
Image Conversion
Image conversion is also part of FFmpegForge.
The application exposes FFmpeg's available image formats and encoding capabilities, allowing users to work with both common and less frequently used image formats.
Freedom of Format and Encoder Selection
One of FFmpegForge's main goals is to give users more control over their conversions.
Rather than deciding in advance which formats and encoders users should have access to, the application lets them explore and select from the capabilities detected in FFmpeg.
This makes FFmpegForge useful not only for everyday multimedia conversions but also for users who need to work with unusual or specialized formats.
Modern Windows Interface
FFmpegForge features a modern dark-themed graphical interface designed to make FFmpeg easier to use without requiring users to manually write command-line instructions.
An Important Note About Format Compatibility
Although FFmpegForge dynamically discovers FFmpeg's available formats and encoders, this does not mean that every possible conversion combination is supported.
Some formats are limited to decoding or encoding, and certain containers only accept specific codecs.
Actual conversion capabilities depend on FFmpeg's technical limitations, the selected formats and encoders, and the FFmpeg build being used.
Why I Created FFmpegForge
My goal was to create a graphical interface that doesn't unnecessarily restrict the capabilities of FFmpeg.
FFmpeg supports an enormous range of multimedia formats and codecs, but many of these capabilities can be difficult to access without using the command line.
With FFmpegForge, I wanted to provide a more accessible way to explore and use these capabilities while maintaining flexibility for users who need more than standard conversion presets.
The project is under active development, and I'm continuing to improve it.
Screenshots
[Insert your FFmpegForge screenshots here]
Download
[Insert your official download or GitHub link here]
Feedback Welcome!
I'd really appreciate feedback from the VideoHelp community, especially from users who are familiar with FFmpeg or other FFmpeg-based graphical interfaces.
I'm particularly interested in your opinions on:
- The dynamic detection of FFmpeg formats and encoders.
- The ability to access uncommon video, audio, and image formats.
- The interface and overall usability.
- Features or improvements you'd like to see in future versions.
If you already use another FFmpeg GUI, I'd also be interested in knowing what you think FFmpegForge could do differently or better.
Constructive criticism, suggestions, and bug reports are all welcome.
Thanks for taking the time to check out FFmpegForge!
https://github.com/Aaronstanley78/FFmpegForge
+ Reply to Thread
Results 1 to 20 of 20
-
Last edited by Aaronstanleyfx; 19th Oct 2026 at 05:43. Reason: More suitable presentation of the software
-
Hi everyone!
I'm excited to announce the release of FFmpeg Forge v1.1.1!
This new version brings several exciting features and improvements for video enthusiasts, content creators, and anyone looking for a powerful yet easy-to-use FFmpeg graphical interface.
FFmpeg Forge is a project I've been developing to make video and audio processing more accessible, without having to deal with complicated command lines, while still offering advanced customization options.
With version 1.1.1, I've introduced NVIDIA RTX AI video enhancement, DLSS Super Resolution, advanced video upscaling algorithms, immersive audio processing, and much more!
Here's a closer look at what's new.
NVIDIA RTX AI Video Enhancement
FFmpeg Forge now supports NVIDIA's AI-powered video enhancement technologies.
RTX Video Super Resolution (VSR)
Improve video quality using AI-powered upscaling and compression-artifact reduction.
Choose between four quality levels, from Low to Ultra, and upscale videos by ×2, ×3, or ×4.
RTX TrueHDR
Convert SDR videos into HDR10 with adjustable contrast, saturation, middle gray, and peak brightness.
The output uses 10-bit video with BT.2020 and PQ metadata for HDR-compatible playback.
DLSS Super Resolution (NVIDIA NGX AI)
FFmpeg Forge also supports NVIDIA DLSS Super Resolution with several processing modes.
DLAA, Quality (×1.5), Balanced (×1.72), Performance (×2), and Ultra Performance (×3).
⚙️ Three RTX Processing Engines
FFmpeg Forge can automatically select between three different processing engines depending on what's available on your system.
1. Built-in NVIDIA Driver Engine
Uses ForgeRTX.exe through D3D11 and NVIDIA driver extensions, similar to the technology used by Chrome, VLC, and mpv.
This engine works directly with compatible NVIDIA RTX GPUs without requiring additional SDK installations.
Simply enable RTX Video Enhancement in the NVIDIA App.
2. FFmpeg NVIDIA RTX Filters
Supports vsr_cuda, vsr_drv_cuda, truehdr_cuda, truehdr_drv_cuda, isr_cuda, and dlpp_drv_cuda, depending on your FFmpeg build.
The visual editor also provides access to the available filter parameters for more advanced customization.
3. RTXVideoProcessor
Optional integration with RTXVideoProcessor, based on NVIDIA's official RTX Video SDK.
Supports VSR ×2 and TrueHDR with NVENC encoding.
You can place RTXVideoProcessor.exe inside C:\FFmpeg\bin\ or locate it manually through the application.
GitHub: https://github.com/DrC0ns0le/RTXVideoProcessor
Advanced Video and Image Upscaling
This version also introduces several additional scaling algorithms for both videos and images.
AMD FSR 1, NVIDIA Image Scaling (NIS), Snapdragon Game Super Resolution (SGSR), Anime4K, xBR, hqx, Integer Scaling, Nearest Neighbor, Sharp Bilinear, BCAS, EWA Lanczos Sharp, Lanczos, Bicubic, and Spline.
GPU acceleration through libplacebo and Vulkan is enabled by default when supported.
The FSR, NIS, and SGSR shaders are embedded directly into FFmpeg Forge, so there's no need to download additional shader files.
A full FFmpeg build with libplacebo support is recommended for these features.
Immersive Audio – Dolby Atmos and DTS:X Speaker Layouts
FFmpeg Forge v1.1.1 introduces advanced multichannel audio configurations.
Available speaker layouts include 5.1.2, 5.1.4, 7.1.2, 7.1.4, 7.1.6, 9.1.4, 9.1.6, and 22.2.
These configurations are available in Audio Settings → Channels.
Audio can be exported as discrete multichannel PCM 24-bit or Opus, supporting up to 24 channels.
Automatic Immersive Audio Upmixing
One of the features I'm particularly excited about is the new automatic immersive audio upmixing system.
FFmpeg Forge can transform stereo audio into a multichannel mix designed for immersive speaker configurations.
Dialogue and lead vocals are directed toward the center speaker.
Instruments are distributed across the front, wide, and surround channels.
Bass and kick drums can be routed to the LFE channel.
Reverb, ambience, cymbals, and spatial details can be distributed across surround and height speakers.
Four processing styles are available: Auto, Music, Cinema, and Live Concert.
The application automatically selects a compatible audio codec when necessary.
This feature creates a channel-based immersive mix rather than native Dolby Atmos object-based audio.
Dolby Atmos and DTS:X Passthrough
FFmpeg Forge can also preserve existing immersive audio tracks without re-encoding them.
Supported passthrough formats include Dolby TrueHD Atmos, Dolby Digital Plus Atmos (E-AC-3 JOC), and DTS:X.
The original audio stream is preserved bit-for-bit when passthrough is used.
Please note that FFmpeg itself cannot encode native Dolby Atmos objects or DTS:X metadata.
Improved Compatibility and Error Handling
I've also improved how FFmpeg Forge handles compatibility issues.
The application now provides clearer messages when no compatible NVIDIA RTX processing engine is available.
It also warns users when certain features require a specific FFmpeg build or when RTX TrueHDR is combined with an incompatible 8-bit codec such as H.264.
⚙️ All the Existing FFmpeg Forge Features Are Still Here!
In addition to these new features, FFmpeg Forge continues to offer a wide range of video and audio processing capabilities.
Support for the output formats and encoders available in your FFmpeg build.
Apple ProRes and Avid DNxHR profiles.
Custom video resolutions without predefined limits.
More than 500 FFmpeg filters accessible through a visual editor.
Advanced encoding options and customizable presets.
Batch processing with drag-and-drop support.
Extensive control over video and audio conversion.
My goal is to offer the flexibility of FFmpeg through a convenient graphical interface, without restricting users to a few basic conversion options.
Requirements for NVIDIA RTX Features
An NVIDIA RTX GPU and a recent NVIDIA driver are required.
RTX Video Enhancement should be enabled in the NVIDIA App for the built-in processing engine.
DLSS Super Resolution requires an FFmpeg build containing NVIDIA NGX filters. If these filters are unavailable, FFmpeg Forge can fall back to RTX Video Super Resolution.
RTX TrueHDR requires a compatible 10-bit output codec such as HEVC, AV1, VP9, ProRes, or DNxHR HQX.
Some advanced scaling algorithms require a full FFmpeg build with libplacebo and Vulkan support.
Installation
Installing FFmpeg Forge is straightforward.
Step 1: Place ffmpeg.exe and ffprobe.exe inside C:\FFmpeg\bin\. A full FFmpeg build is recommended.
Step 2: Extract FFmpegForge.rar into C:\FFmpeg\.
Step 3: Launch FFmpeg Forge and start exploring its features!
I'd Really Love to Hear Your Feedback!
FFmpeg Forge is a project I'm actively developing, and I'd genuinely love to hear what the VideoHelp community thinks about it.
I've put a lot of time and effort into this software, and my goal is to keep improving it with each release.
Whether you're an experienced FFmpeg user, a video encoding enthusiast, or someone simply looking for an easier way to convert and enhance videos, your opinion matters to me.
I'm particularly interested in knowing what you think about the interface, the available features, and the overall user experience.
Do you find FFmpeg Forge easy to use?
Which features do you find the most interesting or useful?
Is there anything you think should be improved or redesigned?
Are there any features you'd like me to add in future updates?
Have you encountered any bugs or compatibility problems?
And for those of you who have used other FFmpeg graphical interfaces, how do you think FFmpeg Forge compares?
Please don't hesitate to share your honest opinions, whether positive or negative.
Constructive criticism is always welcome, and I believe it's one of the best ways to improve a project.
Even if you only try the software for a few minutes, I'd be very interested in hearing your first impressions.
I'd also love to know if FFmpeg Forge is something you could see yourself using regularly, and what would make it more useful for your workflow.
Your suggestions and feedback could directly influence the features and improvements included in future versions.
Thank you to everyone who takes the time to download, test, and share their thoughts about FFmpeg Forge.
I really appreciate your support, and I'm looking forward to hearing what you think!
Enjoy FFmpeg Forge v1.1.1, and happy encoding!
Download new version 1.1.1 here: https://github.com/Aaronstanley78/FFmpegForge/releases/tag/1.1.1 -
Hi.
Interesting software.
How can I use CRF mode for H.265 (nvenc)? -
Update 1.1.3.a
��️ CRF for H.264 and H.265 NVIDIA NVENC: when the video codec is "H.264 GPU NVIDIA (NVENC)" or "H.265 GPU NVIDIA (NVENC)", a new "CRF NVENC" field appears in Video settings. Type any value from 1 to 51, like the CRF of x264 and x265.
1 gives the best quality and the biggest file, 51 the smallest file. Good values are 18 to 23 for H.264 and 20 to 26 for H.265.
0 keeps the automatic mode: the value comes from the Quality setting, as before.
A hint under the field tells you what the chosen value means (very high, high, medium or low quality).
The value is used as NVENC constant quality (FFmpeg rc vbr with cq), so the file size follows the picture, like a real CRF.
It works everywhere FFmpeg Forge encodes with NVENC: normal conversions, Scale (the screen quality option no longer changes your value) and SeedVR2.
A bitrate above 0 still takes priority over the CRF.
Your CRF is saved with your settings.
⚙️ Requirements
An NVIDIA GPU with NVENC (GeForce GTX 10 series or newer) and a recent driver.
�� Download
�� FFmpegForge-1.1.3a-Portable.rar (about 204 MB): everything included, FFmpeg Forge 1.1.3a with FFmpeg 9.0.2 full and ForgeRTX. Extract the whole archive anywhere and run FFmpegForge.exe.
https://github.com/Aaronstanley78/FFmpegForge/releases/tag/1.1.3.a -
Hello, thank you for your feedback.Originally Posted by Zee
This is included in the new update that has just been released. Enjoy!
-
Looks like pure AI slop, even your original post (that whatever AI wrote for you), you couldn't be bothered to fill in the placeholder text it left for you.
Your github repo has obvious AI tells, as does your readme. You don't disclose it, presumably out of shame. And your readme again has broken links and 404 images, couldn't you be bothered to even check what the AI wrote for you? Your Github avatar is AI slop too.
If you can't be bothered to even write your own post about it, and you can't be bothered to write the software itself, why should we be bothered? ��
Assuming it's Claude, why are you calling Claude "I"?
"I" have been developing this, yeah right �� -
Yes, FFmpegForge was entirely coded with Claude. And?
I'm not ashamed of using AI, and I don't see why I should be. I had an idea, guided the development process, requested features and improvements, fixed issues with Claude's help, and personally tested the software before releasing it.
I never claimed to have manually written thousands of lines of code.
If you've found broken links or documentation issues, thanks for pointing them out. Those are valid criticisms, and I'll address them.
But calling someone's work "AI slop" without even discussing how the software actually performs says more about your attitude than about the software itself.
You seem more interested in how the software was created than whether it actually works or is useful.
AI is a tool. Using it doesn't automatically make a project worthless.
Constructive criticism is always welcome. Condescending remarks, not so much.
And if FFmpegForge isn't for you, that's perfectly fine. Nobody is forcing you to use it. -
Wanted to test the effectiveness of the used Optiscaler GLSL script with different settings on non cg content.
Since I could not find a preview option: A preview function with a before<>after (maybe also diff) preview would be helpful when trying to adjust filtering settings.
Cu Selurusers currently on my ignore list: deadrats, Stears555, marcorocchini -
That's a bit aggressive way to comment imho. I;m sure most of the people who are into encoding/conversion wouldn't mind and didn't care if this is result of human or human supported by AI.Originally Posted by Roph
-
Roph, you are sadly misinformed about what AI-guided development is all about. You should do some research before you embarrass yourself further. And as ctsdh0 said, who cares if the application works well.
-
Hi,
First of all, I wanted to say a huge thank you for creating FFmpegForge. I am currently using version 1.1.3a, and it is a truly impressively put-together program! I especially appreciate that it is a proper portable version that works flawlessly no matter where it is placed on the computer. That kind of flexibility is amazing.
Since I am very interested in DSD audio, I have a small feature request. The latest FFmpeg nightly builds now include a native DSD encoder and fully support .dsf and .dff extensions.
Would it be possible to add these two options to the audio encoder dropdown menu in a future update? It would be a fantastic addition for high-end audio processing.
Thanks again for your hard work and for consideration!
Best regards -
Hello everyone, and thank you for your support.
A video preview feature is planned for a future update I am currently working on it. Ivek167, thank you for your suggestion to add a DSD encoder to the "audio" section this will be included in a future update. I’m glad you like the software I am doing my absolute best to create a truly comprehensive FFmpeg-based program. For those with a GitHub account, please consider starring the repository if you support and enjoy the software. It’s a pity that VideoHelp hasn’t added it to their software list I submitted a request, but it still hasn’t appeared there.Last edited by Aaronstanleyfx; 30th Oct 2026 at 01:45.
-
Nothing - however this is good practice to inform community about AI use.Originally Posted by Aaronstanleyfx
Well... you are not ashamed but also seem you are far from being proud of fact AI use...Originally Posted by Aaronstanleyfx
True, you never claimed that you manually written thousands line of code however those thousands line of code was created by using other people work, frequently without permit - LLM's stealing other people work - this well known fact - so you personally not stealing but you are simply taking benefits from thievery of others.Originally Posted by Aaronstanleyfx
Yes, AI is tool, tool build on humans knowledge - those LLM was trained using OTHER people work and companies running those LLM's didn't payed humans single cent for their work, effort and knowledge - that's why AI is thief and this must be pointed clearly.Originally Posted by Aaronstanleyfx
But this is beyond you - you must accept various comments and remarks also those pointing that your development is based on knowledge stolen from other people...Originally Posted by Aaronstanleyfx
-
You can use SoX for DSD, also foobar2k supports DSD...Originally Posted by Ivek167
https://github.com/rossturk/sox-dsd
https://github.com/turbulentie/sox-dsd-win
This seem the most regularly maintained SoX with DSD support for - https://codeberg.org/sox_ng/sox_ng
https://codeberg.org/sox_ng/sox_ng/wiki/DSD-Encoding
It is good to mention that DSD support was possible thanks to HUMAN work - HUMAN behind DSD code is Måns Rullgård not some LLM - LLM's are not aware of Nyquist rate concept...Last edited by pandy; 29th Oct 2026 at 13:49.
-
⚠️ Beta version: please report anything strange in the Issues.
✨ New
Live before / after preview: while a video exports, a preview window opens and shows the original frame and the final rendered frame (Scale, FSR, OptiScaler, Denoiser, filters, RTX, SeedVR2, VisionNext…).
↔️ Slider mode: drag the line (or use the arrow keys) to wipe between the original and the final picture. Side by side mode: both pictures next to each other (Space switches between the two modes).
⏱️ The preview updates about once per second of video. The frames come from the same FFmpeg command as the export, so the video is decoded only once and the export is barely slower.
☑️ Option "Live before / after preview while exporting (beta)" in Video settings, ticked by default. If a file cannot be previewed, it is converted normally without the preview.
DSD export: new audio formats DSF (Sony DSD Stream File) and DFF (Philips DSDIFF), and new audio codecs DSD64 2.8 MHz and DSD128 5.6 MHz.
FFmpeg can read DSD but cannot write it, so FFmpeg Forge does it itself: FFmpeg prepares the audio (trim, channels, volume, filters, oversampling), then FFmpeg Forge's 1 bit 7th order delta sigma modulator writes the DSD stream. Measured THD+N (20 Hz to 20 kHz): about −113 dB in DSD64 and −124 dB in DSD128.
Mono, stereo, 3.0, quad, 5.0 and 5.1 (DSF and DFF hold up to 6 channels). 0 dBFS is written at 50 % modulation, the SACD reference level.
✅ Everything from 1.1.4 is still there: Upscale VisionNext, SeedVR2, OptiScaler, Denoiser, Scale, NVIDIA RTX AI, CRF NVENC, Spatial Music, filters and all formats.
Downloads
FFmpegForge-1.1.5b-Portable.rar (about 204 MB): everything included, FFmpeg Forge 1.1.5b with FFmpeg 9.0.2 full and ForgeRTX. Extract the whole archive anywhere and run FFmpegForge.exe.
FFmpegForge-1.1.5b.rar (about 0.6 MB): normal edition for your own FFmpeg (C:\FFmpeg\bin or next to the .exe).
ℹ️ The live preview needs FFmpeg 4.4 or newer (included in the portable edition).
https://github.com/Aaronstanley78/FFmpegForge/releases/tag/1.1.5b -
Why? Do you inform them about your lint utility or your text editor?Originally Posted by pandy
That's an interesting point. I asked Claude about it:True, you never claimed that you manually written thousands line of code however those thousands line of code was created by using other people work, frequently without permit - LLM's stealing other people work - this well known fact - so you personally not stealing but you are simply taking benefits from thievery of others.
> For coding, do you observe licensing and copyright or not?
● Yes, I try to observe licensing and copyright, though I can't guarantee it every time.
- I avoid reproducing substantial copyrighted material word for word, whether it's code, book text, or lyrics.
- When you bring in third-party code, I keep its license headers and attribution, and I flag license conflicts,
for example GPL code going into a closed-source program.
- The limit is that I have no way to check my output against every existing source, so an occasional memorized
snippet could slip through without my knowing.
Claude says:...your development is based on knowledge stolen from other people...
Where the critics have a real point:
- Anthropic was sued by authors over training data. In June 2025 Judge Alsup ruled that training on lawfully acquired books was fair
use, but that downloading copies from pirate sites was not. Anthropic then agreed to a settlement of about $1.5 billion.
- Consent and payment for the creators whose work was used are a legitimate, unresolved grievance. That's an argument about
what's fair, not just about the law.
Where "theft" and "stolen property" go wrong:
- Copyright infringement isn't theft in law. The US Supreme Court said as much in Dowling v. United States (1985): infringed copies
are not "stolen goods".
- Whether training on copyrighted material is legal is still contested. Early rulings (Alsup in the Anthropic case, Chhabria in
Kadrey v. Meta) found training to be fair use on the cases before them. Other cases are still open, so "it's settled theft" isn't
accurate. Neither is "it's settled legal."
- A user isn't receiving stolen property. Liability for a user would come from publishing output that reproduces someone else's
protected work. That's the same duty you have with code from Stack Overflow, a forum post, or a book.
- Anthropic's commercial terms include copyright indemnity for customers against claims over outputs.
Training on copyrighted work is a live legal question, and early rulings have mostly found it fair use. Where Anthropic used pirated
books, a court said no and Anthropic paid for it. Using the tool doesn't make anyone a receiver of stolen goods. What matters is
whether you publish something that copies someone else's work, and that was true before AI.
Your over-the-top rhetoric is exposing you now. Also many top developers of important and respected apps are using AI coding tools now. There is nothing to be ashamed of.It is good to mention that DSD support was possible thanks to HUMAN work - HUMAN behind DSD code is Måns Rullgård not some LLM - LLM's are not aware of Nyquist rate concept...
Claude says:
Nyquist rate: for a signal band-limited to B Hz (no energy above B), the Nyquist rate is 2B, the minimum sampling rate at which the
signal can be reconstructed exactly from its samples (Nyquist-Shannon sampling theorem). In practice you must sample strictly above
2B: a sinusoid at exactly B can be lost if every sample lands on a zero crossing. Reconstruction is ideal sinc interpolation
(Whittaker-Shannon). Sampling below 2B causes aliasing: components above fs/2 fold back and show up as false lower frequencies, and
once aliased they cannot be separated out. That is why an anti-aliasing low-pass filter goes before the sampler. Do not confuse it
with the Nyquist frequency, which is half a given sampling rate (fs/2), the highest frequency that rate can represent. Example: CD
audio samples at 44.1 kHz to carry a 20 kHz band; the gap between 20 kHz and the 22.05 kHz Nyquist frequency leaves room for the
anti-aliasing filter's transition band.
One more point worth having: the theorem requires the signal to be strictly band-limited. Real signals never
quite are, so practical systems oversample and filter, rather than sampling at exactly 2B.
-----
I suggest you start a new thread for this to avoid further polluting this excellent thread. Finally, you may not have considered this, but if you go around accusing people of receiving stolen property you open yourself to defamation lawsuits.Last edited by Columbo; 29th Oct 2026 at 16:49.
-
That was Ivek167.Selur, thank you for your suggestion to add a DSD encoder ..
about the preview: I had more a filter preview in mind (maybe by leveraging ffplay or libav), but thanks for the 'encoding viewer / comparer'.
Cu Selurusers currently on my ignore list: deadrats, Stears555, marcorocchini -
Poor argument - ask LLM's why - ask about difference between text editor and LLM (i give you a hint - syntax highlighting is not AI).Originally Posted by Columbo
Firstly - licenses usually not covering LLM training - they are tailored for human use not machine use, secondly LLM's are guided to try not to do some things but as we already know they also frequently ignore those guidelines and do forbidden things - recently we had large number of event when LLM violate law and hacked security of services to acquire data - that's why there is informal rush to create better protection also in laws against LLM.Originally Posted by Columbo
Law and knowledge are lagging when compared to LLM's evolution speed.
How many of people creating something innovative is aware of fact that their work was used illegally by LLM? How many of them is able to fight against companies running LLM's?Originally Posted by Columbo
So you generally quoting that indeed LLM's stealing other people work but if those people are reach and they can afford for lawsuit against big tech companies they can be payed for their work - good for you.
Good academic discussion only in wrong place - people feel there work was stolen, quoted by you LLM says about morality vs law and we all knows that LLM's are Super Intelligence due not using moral rules in their decisions - they are above human morality.Originally Posted by Columbo
So what exactly exposing my "over-the-top rhetoric"?Originally Posted by Columbo
And quoting wikipedia formulas quoted and recitied by LLM doesn't mean that LLM i AWARE of CONCEPT behind formula - i saw LLM code when LLM proposed generation of sinusoid 200kHz and used sampling 44100Hz, i saw multiple times when LLM's was unable to do simple ESTIMATION based on quite precision input data - some LLM's hallucination are even hilarious but still - this is pure syntax shuffling.
So once again pure argumentation mostly limited to providing quotes from LLM that prove my point.
You are barking wrong tree but this is your circus and not my monkeys - if you feel insulted then sue me - your LLM may prepare lawsuit for you - waiting patiently... good luck.Originally Posted by Columbo
BTW Many forums and sites introduce rules to clearly mark AI generated content, some sites forbid or restrict such content - perhaps it is time for Videohelp to introduce new rules addressing AI? -
✨ New in 1.1.5
Upscale VisionNext without Visual Enhancer: VisionNext is now built in. Its model is VOSR 2.0 (one step 1.4B generative super resolution, CVPR 2026, Apache 2.0). Click "Install VisionNext model…" once and FFmpeg Forge installs the official VOSR code, the VOSR 2.0 weights with their VAE and DINOv2 encoder, Python and PyTorch CUDA in its own folder (VisionNext next to FFmpegForge.exe), then runs the model itself on your NVIDIA GPU.
GPU or CPU: tick GPU (NVIDIA CUDA, fast) or CPU (works without an NVIDIA card, much slower). The choice appears as soon as Upscale VisionNext is ticked.
Upscale factor ×1, ×2, ×3, ×4 and ×8, with tile size 256 or 512. The work grows with the output size: ×3 takes about 2 times longer than ×2, ×4 about 4 times and ×8 about 16 times. Very large outputs switch the VAE to tiles automatically so the GPU memory is not exceeded.
Videos are encoded directly, never picture by picture. With the built in engine, the video goes from the decoder to the model and straight into the final encoder: no picture files on disk and one single encode. With the Visual Enhancer engine, whose command line only reads pictures, FFmpeg Forge feeds it small batches of frames while the final encoder is already running, so there is still one single encode and no full frame sequence on disk.
⚡ Faster VisionNext: bf16 on the tensor cores with flash attention, all the tiles of a frame processed in one pass, and the same noise for every frame so there is no flicker between frames. Norms, tile blending and the colour fix stay in full precision. Triton is no longer needed (it does not exist on Windows), which fixes the "TritonMissing" error.
Denoiser IA: the Denoiser section is now a single checkbox. Tick it and FFmpeg Forge removes noise and film grain from videos and images with an AI network, for a clean and natural picture in the spirit of Neat Video. The AI is SCUNet (Swin Conv UNet by Kai Zhang et al., Apache 2.0), trained for real world noise: camera sensor noise, film grain and compression noise. A Neat Video style temporal stabilisation keeps the picture free of flicker.
Much faster Denoiser IA: the whole SCUNet network is captured once as a CUDA Graph and replayed in a single call per batch, instead of launching about 8,500 small GPU operations one by one. The window attention is fused into one GPU kernel, up to 8 frames are processed per pass, in fp16, with decoding, AI work and encoding running at the same time. If the GPU memory is full, the work size shrinks automatically, and the normal mode is used if CUDA Graphs are not available.
Live preview: the before / after window closes by itself when the export is finished, and it can no longer make an export fail. If its pictures cannot be written (TEMP folder cleaned by Windows or an antivirus, file locked), only the preview stops and the export goes on.
Fixes: clearer error messages for the built in VisionNext engine, a warning when the output is too large for H.264 NVENC (4096 px) or H.265 NVENC (8192 px), and picture sizes that are not a multiple of 8 now work with VisionNext.
✅ Everything from the 1.1.5 betas is included: live before / after preview, DSF and DFF in every channel layout (import and export), SeedVR2, OptiScaler, Scale, NVIDIA RTX AI, CRF NVENC, Spatial Music, filters and all formats.
⚙️ Requirements for the AI features
An NVIDIA GPU (CUDA) with a recent driver is recommended. Without one, the AI runs on the CPU, which is very slow.
About 3 GB of disk space for the Denoiser IA, and about 10 GB for VisionNext built in (24 GB of VRAM recommended, more for ×4 and ×8).
×3, ×4 and ×8 need the built in engine. For outputs larger than 4096 px choose libx264, libx265 or H.265.
Downloads
FFmpegForge-1.1.5-Portable.rar (about 206 MB): everything included, FFmpeg Forge 1.1.5 with FFmpeg 9.0.2 full and ForgeRTX. Extract the whole archive anywhere and run FFmpegForge.exe. The AI Denoiser and VisionNext install next to it.
https://github.com/Aaronstanley78/FFmpegForge/releases/tag/1.1.5
Quote
Visit Homepage
