Disable "Nodes 2.0" in your comfyUI settings to see the sliders. Toggle for Finish Mode will look like it's disconnected, because it doesn't take any links. It just bypasses the matching color group, which is the Black group underneath it titled "Finish Mode"

@louiscXXX ComfyUI-mxToolkit is the name of the nodepack that has those sliders. Make sure you have 0.9.92 installed correctly. that's the only other reason I can think they might not show up.

I believe that occurs if you don't have an image or video loaded into either your start frame input, or (more likely) your end frame input. This is a FFLF only workflow, meaning you can't just use a single image and gen from it. It's designed to take a last frame input of either an image or video, so make sure one is loaded and that you have it toggled to be enabled. If you want a single frame input workflow, I have one available, just go to my profile and you'll see it; it's the only other workflow I've posted.

Oh yes, color-space drift is always a thing in any model. Typically, as I show in the video, I go with 33 frames of overlap so I have ample room in to blend the 1st gen into the 2nd gen in my video editor, Davinci Resolve. I also use color adjustment settings to match the two together more accurately. Usually the color drift is easy to compensate for. So much so that I barely even consider it a hindrance or issue.

Ahh, I think I may have some answers for you. Let's say your Input video is 1920x1080 and so you've set your target resolution to 1920x1080 as well. This workflow subdivides by 2, and then by 2 again, making your stage 1 videos 480x270 resolution right? Well, no, not so. 270 is not a valid resolution length to LTX, so it will go to the nearest acceptable resolution, 256 I believe. So then it goes to stage 2 and you get 480*2x256*2 = 860x512. Stage 3? 1920x1024. Notice it's 1024, not 1080 anymore.

SO! If your input video for the extension IS actually 1920x1080, your extension comes out as 1920x1024, which when stretched vertically maintaining aspect ratio to match 1080, means the video will be zoomed in the difference of those pixels, a roughly 3% difference or so.

It is a limitation in LTX multi-stage workflows that I haven't found any solution for. It's all fine as long as all your videos are the actual same resolution. The zoom only occurs with differing video resolutions. That said, you can zoom in your 1st gen the same 3%~ or so as your 2nd gen to compensate. It's just annoying to have to.

@boobkake22 Interesting idea. The kinds of things you gotta start thinking about when releasing a workflow to the public, haha.

@boobkake22 Right back at you! <3

Ah yea, it's a big topic, definitely warrants a video. The short of it:
1. I converted an image of princess peach into a "real life" style using Klein 9b Image Edit Model
2. I used that same model to put her into a park (just load the reference pic of her we just got, set to wide screen latent size, and prompt "Generate a high quality photograph of the girl from image1 standing in a pretty park on a sunny day", as an example)
3. Again with Klein 9b Image Edit, load that pic of her in the park we just created, with a new prompt: "Generate a high quality close-up photograph of the woman, she is looking at the camera with a subtle amused expression," etc. etc.

This is how I made the keyframes for that cold opening sequence, more or less. There is definitely more to the process than that, but hopefully that gets you started.

Comfy thinks SOMETHING has changed for it to run through the old stages again. It's programmed to avoid repeating work it's already done, and yea I can see how annoying it would be to have to regen them to go to stage2. I suppose it's possible there is some setting or configuration in comfy that's different between your install and mine that allows it to redo existing work on runs. I wish I could look at your wf/setup and make sure nothing got switched. Like if the Stage 1 "Manual Random Seed" node isn't set to fixed or something.

@jackaroo1432111 This is super helpful information! Thank you for your edits. I was definitely scratching my head, because I know of one other person experiencing what you had, but they were always experienced comfy users and not complete newbies like you'd expect. Which makes sense considering complete newbies don't know how to pass startup flags like turning cache off LOL. Makes perfect sense! Also I am glad it's working well now!

TBH I didn't use end video much before I released this workflow, but I did use it extensively the other night and noticed the same results. LTX is hyper-fixating on the first frame of the end video. It treats the 'end video' as "transition to the 1st frame of the end video, and then play the video." I was not smart enough to figure out a way to get LTX to consider all the frames. So it's one of those situations where, I know what the problem is, but not how to implement the solution.

@drowai443 I will take a look. Obviously my workflow can extend videos seamlessly with no problem, but I agree that the end frames being video leads to often problematic transition results. I'd like to fix it since it's a cool feature. The idea that the gen in between two videos is just the transition between those two videos is a valuable concept.