All things migrate, universal sticker，Support nunchaku
Workflows:https://www.runninghub.ai/post/1955386628840701953?inviteCode=rh-v1172 Register for free use
https://civarchive.com/models/1793451?modelVersionId=2029628
If you find it quite good or have any questions, please send me more pictures to help me grow. Thank you for your participation. Please try to give me the kind with a white background sticker
Modification: Version V.1, hand distortion issue, automatic light tracking, automatic background fusion (see auxiliary prompt words)，Transparency of transparent and semi-transparent items
New features: Tattoo: (This pattern is tattooed on the arm), Print: (This pattern is printed on clothes), Furniture: (This sofa is placed in the corner), Face swap: (This woman is looking at the audience), Dress change: (The woman is wearing this dress), Jewelry: (This necklace is worn around the neck), Item: (This bottle of perfume is placed on the table), Shoe change: (Both feet are wearing this shoe), Remove, Expand image: (Fill in the blank area)，The character in the sticker is obedient: (This cat stretches out its hands to make a heart gesture and opens its mouth to smile)
The texture must be a white background image. Just place it where you want
Trigger word: Put it here Add auxiliary prompt words：Adaptive, automatic light adaptation, 4K, high definition, outstanding picture quality
Note: The best effect is achieved with a white background and white borders all around!
Workflows:https://www.runninghub.ai/post/1955386628840701953?inviteCode=rh-v1172 Register for free use
https://civarchive.com/models/1793451?modelVersionId=2029628
We recommend a matting plugin to help your images fit better and maintain consistency
https://github.com/XiaoHeiziGGG/put-tools/tree/main?tab=readme-ov-file
https://github.com/zml-w/ComfyUI-ZML-Image
If you need commercial customization or have any suggestions or assistance, please contact me
Description
FAQ
Comments (65)
MUST have!!!!
Haha, thank you, bro.
好顶赞
Thank you for your support. We welcome more suggestions. :）
Looks fantastic
Thank you for your support. We welcome more suggestions. :）
prompt examples?
Sorry. What do you mean?
Futurlunatic what we need to write to achive this result?
Oh oh. In the prompt word introduction, there is "put it here" + the simple prompt word given in the previous example. Don't be too complicated, as it may mislead the AI
altolstyh386 Oh oh. In the prompt word introduction, there is "put it here" + the simple prompt word given in the previous example. Don't be too complicated, as it may mislead the AI
Even without the LoRA, Kontext has this ability (it is amazing, when you think about it, that models release WITHOUT an instruction manual, and users have to experiment to discover what a model can do). However, as with all model abilities, LoRA finetuning makes them better and more reliable.
The refined sticker method is clearly a sort of next gen inpainting.
Thank you for your comment. After extensive testing by dozens of people, the effect of adding this Lora in kontext will be much better. It will better maintain consistency and the responsiveness of prompt words. If you like, you can do a job locally to test it out. There will be surprises you never expect. Thank you again for giving me your evaluation. thank you
In fact, kontext itself is certainly generalizable in terms of face and head swapping, but its capabilities are insufficient. This LoRA clearly strengthens this concept and specifically optimizes the effects of lighting, shadow, and subject integration with the background.
2514849841630 Thank you for discovering these capabilities. Although there are some shortcomings, I will do my best to optimize them in subsequent versions. If you have any questions, please feel free to raise them at any time. Thank you!
don't know why but i can't get it to work with face swaps. my prompt is "put it here. she is looking into the camera."
"Put it here, This woman is looking at the audience, Adaptive, automatic light adaptation, 4K, high definition, outstanding picture quality"
Try
goldenalexeysky727 Great! You've already grasped the essence. Hahaha
Is there special node to combine two images or do you use outside program to do this?
Also it doesnt work with nunchaku ;/ just basic flux kontext
Comfyui-LayerForge
1 node - LayerForge is an advanced canvas node for ComfyUI, providing a Photoshop-like layer-based editing experience directly within your workflow. It extends the concept of a simple canvas with multi-layer support, masking, blend modes, precise transformations, and seamless integration with other nodes
goldenalexeysky727 Good idea
https://github.com/XiaoHeiziGGG/put-tools/tree/main?tab=readme-ov-file
This node can also be very convenient for handling irregular shapes, and the effect will be better
"Guidance" should be "2.0". Only then will there be a correct result. If you specify above, then it will stop placing the object
OK. I've learned...." Although I don't quite understand what you're saying. It's because my English is too weak. Thank you very much for helping to answer the question.
Your words are translated in a mess by the translation software... You know
The flux guidance? Or the sampler guidance?
KVTM It seems that what he said should be right“flux Guidance"
I'll give it a try too. Thank you very much
Попробую на русском. "FLUX Guidance" должен быть "2.0". Если указать значение больше этого - то LORA прекращает работать, не вписывает верхнее изображение с белым фоном в общее изображение
KVTM "flux guidance", yes
goldenalexeysky727 I really want to discuss it directly. Hahaha, I'll try to see what you mean roughly
Futurlunatic https://ibb.co/CKM1zdNB
goldenalexeysky727 Hahaha, I'm so sorry for your trouble. Are you interested in visiting the Chinese community
This is a really useful LoRA! I’d like to ask what you used to train it? Is there any tutorial for training a Kontext LoRA?
v0.1 version much better then v0.2. I believe v0.3 will be much better
First of all, thank you very much for using this LORA. To answer your question: Describe what you want the final image to look like, there's no need to write the process, just the result is fine. Is that clear?
This lora is amazing. You can also say that its the way to force Kontext to maintain a background image. Normally it mixes everything more than the desired. Tried it with multi-chained reference latents, image stitch, manual position of the sticker and the sticker attached to the main image side too. Works better than the regular kontext.
Thank you for your recognition. Also, thanks to the power of the original kontext. I'm just using one way to make kontext better recognize the items to be changed and maintain the consistency between the background and the migrated objects to the greatest extent. If you have any good suggestions, please feel free to put them forward and I'll make improvements
Great work, How many images you used for training?
More than 300 sets of pictures were used. More were added in the later versions
Futurlunatic Wow, any plan for the next version?
allengexx is already here..
In previous comments someone mentioned the workflow updated to v4. When will that be available for testing? Thanks!
Soon, after testing, thank you for your attention
可以分享一下该lora的训练集吗？非常希望学习一下大佬的工作！！！
Sorry, you're overthinking it, bro
哥们,直接把底裤扒了给你得了
Outstanding work 👍
* * * Workflow * * *
https://civitai.com/models/1831687/flux1-dev-krea-kontext-nunchaku-my-20sec-workflow?modelVersionId=2107703
Thank you for your support, little beauty
Can u share some training details? Such as training step, learning rate, etc
How did you prepare the dataset? and can you share training deatils ?
Hi i have tried your lora, it sometimes work and yet it sometimes change the subject to a different person, i am using nunchaku version, not too sure if it affect it?
I don't know what your workflow or mask looks like. I'll put out a detailed tutorial when I get the chance. It's currently in preparation. There's also an update version about to be released.
thank you so much :)
老大，为什么我用白边框的时候就不行，用白底就行？
马上就可以了别急
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.