ComfyUI Writes ProRes 4444 Now. The Alpha Channel Is the Point.

ComfyUI v0.44.0 shipped on November 4. Among the changes was a pull request numbered 17463 that added ProRes output. The changelog entry is about three lines. It is the most consequential thing in the release and probably the least interesting to read about.
Here is why. Until now, if you generated a clip in ComfyUI and wanted it in an edit, the path was H.264. You rendered an MP4, imported it, and lived with the consequences: 8-bit color, 4:2:0 chroma, and a codec designed for delivery rather than for further processing. Every generation you handed over was already degraded before the editor touched it. Grade it once and the banding shows up in the sky.
ProRes is an intermediate codec. It exists to be edited, not to be delivered. The difference in a post pipeline is roughly the difference between handing someone a JPEG and handing them a raw file.
How to use it
Two ways, both simple.
In **Save Video**, choose mov as the container. That writes ProRes.
In **Create Video**, there is a new setting called chroma_subsampling. Set it to 444 and you get ProRes 4444. Set it to 422 and you get ProRes 422 HQ. That mapping comes from the code, so treat it as the source of truth rather than the label in a menu.
The release's tests write and read back ProRes 4444 with a transparent alpha channel, which is the part that matters most and the part barely mentioned in the commit message.
One gotcha, also from the code: if you ask for 12-bit, you get 10-bit and a warning in the console. ProRes 4444 supports 12-bit, but this implementation does not, at least not yet. If you need 12-bit for a grade, you will finish elsewhere.
And a note on the encoder. This uses FFmpeg's prores_ks, not Apple's own ProRes encoder. prores_ks is good, widely used, and decodes cleanly in Premiere and Resolve. It is also not byte-identical to what Final Cut writes, which matters only if your pipeline has a validation step comparing encoder output. The files are large either way.
Why alpha changes the workflow
A transparent alpha channel in a generated clip means the clip is a layer, not a plate.
That distinction determines whether AI footage can be composited in a normal pipeline. Without alpha, you either put the generated clip on a full frame and cut around it, or you rotoscope it out by hand, or you shoot it against green and key it, which defeats the purpose since you generated the background too.
With alpha, a generated character can sit over a real shot. A generated product render can be dropped onto a photographed table. A generated smoke element can be layered over a plate at thirty percent opacity. These are completely ordinary things in a compositing workflow and completely impossible when your only output is an opaque MP4.
The catch is that the model has to produce the alpha, and most video models do not. They generate RGB frames because they were trained on footage, and footage has no alpha. So the newly supported output format is ahead of the models that would use it. That is fine and normal. Format support is a prerequisite, and it has now been met.
For anyone working with models that do produce mattes, or with image pipelines that generate cutouts and then animate them, ProRes 4444 is the container that makes the round trip lossless. You generate, you save with alpha, you drop it in Resolve, and the edge is clean because nothing got chroma subsampled on the way through.
The H.264 detour, explained
If you have never worked in post, the H.264 thing sounds like a minor inconvenience. It is not, and the reason is worth two paragraphs.
H.264 in its common form is 8-bit, 4:2:0. That means each pixel has 256 levels per channel, and the color information is stored at a quarter of the resolution of the brightness information. This is a brilliant set of compromises for streaming a video to a phone. It is a terrible basis for further processing, because every operation an editor does, from a curves adjustment to a key, amplifies the quantization error that is already baked in.
ProRes 422 HQ is 10-bit 4:2:2. ProRes 4444 is 10 or 12-bit with full color resolution and, crucially, an alpha channel. The files are five to ten times larger. That is the entire trade.
In a normal pipeline, that overhead is accepted, which is why cameras shoot ProRes or raw and why editors transcode proxies for speed but always conform back to the high-bitrate source. The intermediate codec is the thing that survives the edit.
AI generation broke that assumption because the tools existed outside the edit suite. You generated in a browser or a Python script, you got an MP4, and the pipeline adapted around the limitation by treating generated footage as finished footage rather than as a source. That worked while AI shots were inserts. It stops working the moment a generated shot needs a grade, a key, or a blend.
What this does not fix
Codec support is not color management. Writing ProRes does not tell the receiving application whether the footage is Rec.709 or sRGB or something else with no tag. If the generated clip looks different in Resolve than it did in ComfyUI, the codec is not the problem.
It also does not fix temporal consistency, resolution, or motion. A 720p clip with flickering details is still a 720p clip with flickering details, now in a larger file. ProRes preserves artifacts faithfully. That is its job.
And the 12-bit gap is real for anyone doing heavy grading or VFX pulls. Ten-bit is plenty for most delivery, and it is not plenty for a hero shot that will get pushed two stops.
The bigger point
The interesting thing about this change is that it is boring.
For two years, the AI video conversation has been about model quality: who has the best motion, who handles hands, who keeps the character consistent. That conversation is real and it is not over. But the workflow conversation is where adoption actually happens, and it is made of exactly this kind of change. A codec. A chroma subsampling flag. An alpha channel that survives a round trip.
Nobody writes a thread about a container format. Everybody who has spent an afternoon transcoding generated MP4s before an editor would accept them noticed this one.
ComfyUI makes no claim to being a professional tool, and it is where a very large number of people actually build AI video pipelines, and a node graph is a surprisingly honest place to see what a pipeline really needs. When the graph starts writing ProRes with alpha, it means the people using it have moved from making clips to making shots. That is the transition that matters, and it does not show up in a benchmark.
Related articles
Chinese text in video always turns to mush. Dreamina handed the typesetting to code
Dreamina launched Motion Studio on November 10. It no longer generates frames; the model writes code that lays out text, shapes and pacing. Chinese titles stop blurring, but whether the content is right is still on you.
A community spreadsheet of taps that turns off the ads inside Chinese Android builds
A no-root tool that switches off the ads Chinese phone makers ship inside the system UI. Its method is a hand-written tap script per phone model, maintained by volunteers every time a ROM moves. It works, and the maintenance is the point.
GPT Image 2.6 held the token price and raised the bill
GPT Image 2.6 kept every token rate from 2.5 and still costs more per image, because more of the work now happens in output tokens. Eleven days into the ramp, the pricing structure says more than the leaderboard does.
Someone rebuilt IDM from the outside and wrote down what they did not touch
A Rust reimplementation of Internet Download Manager claims 5x the speed and 86 languages. The more interesting document is the paragraph explaining how it was built without copying anything.