This is a test workflow that allows you to use the IL\Noob model as Flux.Kontext\Qwen_Edit to generate a character image from a Reference. It works best with the same model as Reference. You also need to add a character prompt.
Attention! The result depends greatly on the seed, so experiment.
I really need feedback and advice on how to improve this! So if anyone is interested, please share your thoughts on this.
Description
1. Workflow simplification.
2. Improved visual workflow structure.
3. Minor control enhancements.
FAQ
Comments (23)
Does this work for automatic1111?
Please don't use "everything everywhere" nodes in complex workflows that you intend to publish. It is brittle and tends to break connections on import -- and then you have no clue where to connect what.
I updated the archive, now there is a version without "everything everywhere"
@Slavikson Thank you!
@morafermi643 In the new version, I tried to make it clearer. Do you think it's better or not?
Thanks for the super cool workflow! I have been using it, but have yet to be able to get any good consistent results do you have any advice?
1. The character description can be very short so as not to oversaturate the model with information, let it guess what things need to be copied.
2. It is better to use a full-length image of the character as a guide. If you need to generate close-up angles, click the toggles.
3. Experiment with different seed and prompt, the results have strong differences.
Thanks all the images I was trying are full length. well mess around more with different prompts. is their a seed you would recommend?
Been trying it out with the example provided prompt/character. It just results in a portrait with a different facial expression. Can you explain what is going on here so we can figure out a good use of this?
Specify the character prompt in "Char Prompt". Specify the pose\action\emotion in "Custom pose". Switch the "full body | upper body" toggles according to your needs.
Where can I download the clip-vision_vit-g clip file ?
Thanks a lot !
Damn, everything is so overloaded. 🙄
The main part on the left, the green part on the right is just an upscale, albeit with some nuances. In the near future I will release version 0.4 where the settings will be clearer. (And some other surprises)
@Slavikson The nuance is that almost 16 new modules are connected there. And each one needs to be read and understood in terms of its functionality.
DetailerForEachDebugPipe
'DifferentialDiffusion' object has no attribute 'execute'
downloaded all the models and this pops up
I'm not sure but try another version in the archive, called "Consistency characters V0.3 not everything everywhere", it might solve the problem. In future versions it won't be there by default.
i had the same problem, its solved by updating comfyui, or just copy paste the diffused diffusion from comfy extras thats in the github