Thanks mate! Maybe we can have a similar workflow using sdxl instantface? It usually produces better results than reactor.
@FiveBelowFiveUK
27 күн бұрын
*spoilers* there is an SDXL pack about to arrive - SODA SDXL, it will share the features KODA and FODA have, InstantID is in that collection - watch this space !
@pixelcounter506
28 күн бұрын
I guess the swapping starts with the most upper face (index0), then the second most upper image (index1) = top-down-approach. If the faces are on the exact same level, then it might go with the index numbering from left to right (needs to be checked). And debugging is always a good idea. As always thank you very much for your published workflow... really appreciated!
@FiveBelowFiveUK
27 күн бұрын
I have the same feeling on this, however - as a viewer has since ponted out, there is an additional options node i was unaware of, which was added to the node release. This should give us control over the ordering and i'll return once that update is proven :)
@wolpumba4099
27 күн бұрын
*Summary* * *(**0:18**)* *Multi-Person Face Swapping:* This video demonstrates a workflow in ComfyUI to swap faces of multiple people in an image using pre-trained face models. * *(**0:18**)* *Workflow Pack V5 (Loki Live):* This pack contains the workflows shown in the video and can be downloaded from Civitai. * *(**0:18**)* *Two and Three Faces:* Examples are shown for swapping two and three faces, but the technique can be extended to more. * *(**0:18**)* *Face Models:* Using pre-trained face models (created with tools also included in the pack) makes the process significantly faster. * *(**1:01**)* *Input Face Index:* The most important setting is the "Input Face Index," which determines which face is replaced with which model. The numbering order can be unpredictable and might require experimentation. * *(**3:02**)* *Gender Detection:* While there's an option to detect gender, it's not always reliable and manual index assignment might be needed. * *(**13:18**)* *Workflow Customization:* The workflow can be customized for more faces by duplicating and connecting relevant nodes, and setting the appropriate input face index for each. * *(**6:06**)* *Image Source:* The workflow can be used with images generated from any AI model. * *(**11:40**)* *Additional Features:* The workflow includes features for masking and in-painting to help with refining the final output. * *(**18:04**)* *Workflow is Modular:* This workflow is designed to be modular, taking an image as input and outputting a modified image, making it compatible with other workflows and ComfyUI setups. Summarized by AI model: gemini-1.5-pro-exp-0801 Cost (if I didn't use the free tier): $0.0588 Input tokens: 14728 Output tokens: 687
@FiveBelowFiveUK
27 күн бұрын
now you just served me good :) This is W summary. did you just give the KZitem link to gemini? i might start doing this.
@wolpumba4099
27 күн бұрын
@@FiveBelowFiveUK that would be against youtube's TOC (if i understand them correctly) i manually copy and paste the transcript. i made a screencast a while ago, you can see in my channel. in the video you can see the ip of ther server i used for testing. the server is still up, if you want you can try (i don't advertise it because i only have 50 pro requests per day). it's basically just sending two requests to gemini to get the summary and then subtitles.
@squallseeker-i2i
28 күн бұрын
You don't seem to be using the Fast Face Swap [Options] node, which attaches to a separate ReActor Options node which allows you to set desired source and input order (e.g. smallest to largest, top to bottom, etc.). I'm only half-way through the vid, maybe it changes by the end.
@squallseeker-i2i
28 күн бұрын
Detect gender doesn't work well enough to bother with it yet. Also while I don't know a way with individual face models to do multiple swaps, you can have a composite of multiple source faces to merge into multiple destination faces using just commas between the index values... so a single node set can change as many faces as you desire. The down-side of source image compared to model source is that it is slower to extract the faces every time than reading the model (thus not as good for video or animations with a large number of swaps involved).
@FiveBelowFiveUK
27 күн бұрын
interesting ! this was not a feature when we last looked at reactor - thanks for the tip ! when i used it before it was a simple left to right from 0. I'm always improving these workflows, so we all pool our knowledge to get there :)
@FiveBelowFiveUK
27 күн бұрын
Yes - i'm a strong believer in the face model route and encourage people to make the face model - my tests showed it's 10x faster and of course, when trained on multi-angle images, more accurate too.
@Gibonsli
27 күн бұрын
I'm looking for a workflow that can swap the face with my character who is a humanoid cat. Is there such a thing?
@FiveBelowFiveUK
27 күн бұрын
interestingly, face swap is optimized for human faces, so you might use inpainting instead, with controlnet of you cats face, align to the same place the head is, hard to do, but possible. might be worth training a lora of your cat, then prompting it in the inpainting. "head swap" with masking. these workflows are in the pack already :) check you have the latest version :)
@DJVARAO
28 күн бұрын
Any face swapper for Flux?
@shvrdavid
28 күн бұрын
This uses an image input, therefore it will work with any model. He even stated that in the video....
@squallseeker-i2i
28 күн бұрын
reactor doesn't care which model type is the source
@FiveBelowFiveUK
27 күн бұрын
The best thing about the loki pack is you can use it with any model, it has an image input and then used it's own models (inswapper8 or mediapipe) - that means you can use it with your flux generated images :) I also think that the outputs are improved with using Flux-Enhance or Flux-Boosted-Upscaler to gain small fixes and detail depth increase after you finish the swapping !
Пікірлер: 17