Hi everyone!
I just finalized a clean, intuitive, and fully modular ComfyUI workflow to generate production-ready character sheets.
Just choose an image, set up what you want, and go — everything is fully automated!
New UPDATE v1.0b
Fixes
- Fixed Save Each Frame ON/OFF
Improved
- Supersampling quality
- Final resize quality
Environment & Props Sheet Generator: Currently in development! It will cover full landscape scenes (Exterior + Interior) alongside item/prop turnarounds. Stay tuned!
Key Features:
Lightning Fast & Standalone: Powered exclusively by MiniMax H3—no extra checkpoints, ControlNets, or external models required.
Flexible Export: Automatically composites the final grid sheet OR/AND directly exports individual raw frames so you can compose your own custom layout.
8 Precise Angles & Shots: Full-body (Front, Back, Side profiles), 3/4 Views (Front & Rear), and 3 Headshot Close-ups with distinct facial expressions.
Style Versatility: Fully customizable—seamlessly handles Photorealism, 3D Renders, Manhwa/Anime, Stylized Concept Art, and more.
1-Pass or 2-Pass Latent Upscale: Run it blisteringly fast in a single pass, or toggle the optional Latent Upscale pass for extra texture refinement.
Modular & Clean: Zero canvas clutter. Built-in padding logic for distortion-free alignment.
Feel free to check it out and let me know your thoughts!
Description
UPDATE v1.0b Changes
Fixes
- Fixed Save Each Frame ON/OFF
Improved
- Supersampling quality
- Final resize quality
FAQ
Comments (23)
This is pretty cool! But for real people i always get the oily look. For all other character types this is really nice 👍️
Hi,
Yes, this issue come with Turbo Lora on Minimax.
In theory to fix it:
You need to download H3 Realism People and put it just after the "H3 reference LoRA". Edit Reference Sheet Prompt Maker and Add at Before "subject_definitions:" write "r34l1sm, "
I will add this fix on next version.
Hope this helps :)
I will see that on next version
Looks like a lot of promise here just from my early testing. I added a second reference image for just a head shot, and it got incorporated into the workflow perfectly. The videos I've created so far in ref2va are showing very good character consistency. In regards to the oily look in realistic style, I'm using the XUELUO ref2va checkpoint along with https://civitai.red/models/2884206/minimax-h3-photorealistic-image-generator-lora-workflow?modelVersionId=3260165 lora set to weight 1.0 in your workflow and it looks real solid. Great work.
Thanks a lot! I'll integrate a slot for this LoRA in the workflow. Thanks again for the tip!
how to added a second reference image?
@wdptt13444 Place another Load Image node (just above the current one is fine) and connect that to a SetNode like the original Load Image node is (name the constant whatever you want, I just made it RefSubject_2)
Next go to the Reference Sheet Prompt Maker node in the Setup section and click on the upper right square to go into the subgraph. Once in subgraph, create a GetNode and click the constant and whatever you named the SetNode earlier should be one of the options, so click that one. Then just connect the GetNode to ref_image_1 input on the MiniMax H3 Reference to Video node.
@wdptt13444 I will add this feature in the next version, which is scheduled for release this week.
@Tacobot As an alternative you can wire the "ref_image_1" input on the minimax REF2V node to the empty subgraph input and put the get node outside of the subgraph.
@SpaceGremlin Would be cool to be able to use all 9 input images. optionally of course.
@hardwire666 👍
@Tacobot I was thinking about what you said in your previous post (while I was adding the second reference field for the next update). Adding a second reference isn't enough; if the prompt isn't modified, the second reference won't be taken into account. Only the first image will be used.
To fix, that in the current context, you need to modify that :
subject_definitions:
@image1 is <Subject 1>, Protagonist shown in <Picture 1> and <Picture 2>.
For those asking for more than two references, I don't understand. Could you explain? In Tacobot's case, I can understand (body + face). But beyond that, I don't see it.
@SpaceGremlin H3 can still take the other images into account. It will just be used less deterministically if not told what to do with it. Meaning the results might not be what you expect or better than you expect, It's kind of a dice roll. However the randomness seems to be working in my favor.
Another modification I've made is for it to save the Images used for the Character sheet, and the all frames to separate directories.
@SpaceGremlin as @hardwire666 mentioned in the next post it can be a bit random adding another image without context but MMH3 is pretty intuitive and I haven't seen an issue when the head shot is similar to the face on the full body shot original. That said, I have added a Text (Multiline) node outside the subgraph and connected it to subject_definitions so I can change them if need. I did the same with the detailed_descriptions so I can change the shots in the character sheet, which I've changed to a 2x3 that's a bit larger. Took out the original image out of the character sheet and now use, full body front, full body back, full body profile, one head shot, and 2 wildcard spots. The wild card spots can be useful for small things you want good detail on, be it a tattoo, a scar...or...other things 😁. That's where a 3rd or 4th reference image comes in handy. MMH3 will get the color and the shape of a tattoo on the back of someone's hand moderately well from a full body image, but you add a reference image of a close-up of that tattoo and include a close-up in your character sheet (one of the wildcards I mention), and MMH3 nails the tattoo at distance, close-up, and in movement.
@Tacobot Haha, glad you like it… Wait until you see the next version which will be released at the end of this week or next week, I've planned lots of features that I think you'll like ;)
More cells, more shooting types, style presets, 1 or 2 native references, preview of images to choose from for each cell...
@hardwire666 "Another modification I've made is for it to save the Images used for the Character sheet, and the all frames to separate directories."
You do not need to make this modification; it is included since begining. You simply need to enable the "Save each Frame" option. The folder containing all the images has the same name as the "Subject Name." ;)
See you soon for the next version.
Where can I adjust final resolution? full body pics on sheet looks low res
Forget it, im blind
@Carpanta85 Haha, you can also extract each frame and put the sheet together yourself using Photoshop or another program. Just enable the "Save Each Frames" option. Each image will be exported to your drive at full resolution (1 MP by default).
Wow this was exactly what I needed an searched for and it works absolutely perfect so a really big THANK YOU!!! Question: Can I somehow choose a shot of the 8 I want? Like closer view of her hand?
"Can I somehow choose a shot"
I am currently working on the next version, which will make that and much more.
Stay tuned ;)
@SpaceGremlin omg cant wait 🥰😍
Working great, Thank you! but can it support multiple references?
Hello and thank you very much.
The next version will allow for two references.
More references are not currently part of the roadmap for the next version. Many new features are in development for V2.
I'm working on two other workflows (Landscape and Props).
I will implement more than two references after that :)