Anima can edit images without inpainting. (In case anyone missed it)
custom node: "https://github.com/Mirumo0u0/ComfyUI-Cosmos-Reference"
example output:
another thing: the workflow is already embedded in the example output, so you can just drag and drop it into ComfyUI. Also, the 'LayerUtility: ImageScaleByAspectRatio V2' and 'Image Compare (RMBG)' nodes are optional—you can just delete them. Inside the workflow, there are some quick notes and the LoRA link included.
Thanks for mentioning this, because so far all of my attempts at editing via inpainting have sucked. I'm sure it works great for other people, just not me, lol.
And thanks especially for showing an example and including a sample workflow.
Is this node effective for masked image generation? Specifically, when using a painted mask for local upscaling and inpainting , do I need to include the surrounding context in my prompt, or should I describe only the content within the masked area?
Is this node effective for masked image generation? Specifically, when using a painted mask for local upscaling and inpainting , do I need to include the surrounding context in my prompt, or should I describe only the content within the masked area?
Yes, masks work, but I'm not sure about upscaling since it's not actually my custom node—I just shared it. For the prompt, just enter what you want to change (use tag-based). Like in my example, the input character is wearing a shirt, and I wanted to change it to a red kimono, so I put 'red kimono' + other details in the prompt. The mask prompt is the same.
Character and style loras worked with AnimaEdit?
How can we guess the prompt which is desired to reference image or, can I put arbitrary prompt for edit?
How can we guess the prompt which is desired to reference image or, can I put arbitrary prompt for edit?
arbitrary prompt directly.
Thanks for this. Very interesting workflow.
Masking and upscaling does work, but this approach sometimes produces seams. I've gotten it to work where I don't notice the seams, it looks to be a limitation of the edit lora and whether or not it decides to radically change the edges of the image. I think with a large enough mask it's mostly ok.
Workflow metadata embedded:
First workflow doesn't seem to work for me, any idea why? I only changed input image.
Ah, I think I figured out, it's a regression in https://github.com/Mirumo0u0/ComfyUI-Cosmos-Reference node
@Nuke1229 Can you please share commit hash of this node you're using?
Ah, I think I figured out, it's a regression in https://github.com/Mirumo0u0/ComfyUI-Cosmos-Reference node
@Nuke1229 Can you please share commit hash of this node you're using?
I'm using the commit right before 'Commits on Jun 1, 2026 Rewrite ApplyCosmosReferenceLatent node'. Alternatively, you can just use this custom node: https://github.com/Anzhc/Anzhc-ComfyUI-Cosmos-Reference