Flux 3 Klein: The Best Way to Generate Character Sheets, by James Doss AI on YouTube. 13 min 45 s.
- Video length
- 13 min 45 s
- Time to do it
- 45 minutes
- Difficulty
- Medium
- Steps
- 10
What you need
Steps
Download the FLUX 3 Image Klein distilled INT8 model files and check the model license before use. Make sure you have access to all three required model components.
Update ComfyUI to v0.41.1 or newer, then load the official distilled image-editing template workflow. Confirm all three model components load without errors.
Pick a clear starting character image with a visible face and simple background. Load it into the workflow as the input portrait.
Write a full-caption prompt describing the character sheet you want, including body views, clothing and style. Expand the caption fully before generation; the optional prompt upsampler is not needed.
Set the seed and generation settings, then run the workflow to produce a full-body turnaround sheet. Keep the seed fixed while you adjust prompts.
Inspect the turnaround sheet for wrong proportions or missing views. If a view is off, correct the prompt or seed and regenerate.
Run a second pass to create a face-angle reference sheet. If the profile view looks wrong, adjust the prompt and regenerate until the angles match.
Generate an expressions sheet with the same character so you have a range of faces for reuse.
Review all results and keep only the references that look consistent. Save them into an organized character reference pack folder.
Test reuse by feeding one saved reference back in to create a new portrait. Check the face, hair and clothing match the original character.
Tips and common mistakes
Keep the seed fixed while you tweak prompts so you can compare changes fairly.
If you hit out-of-memory errors, lower resolution or batch size and close other GPU apps.
Save only clean, consistent images as references; a bad reference will carry errors into new portraits.
Read the model license before using generated characters in commercial projects.
Questions
Do I need to train a LoRA for my character?
No. This workflow uses reference images and prompts instead of LoRA training, so you can reuse a character without training.
What GPU do I need?
The tested setup used an NVIDIA RTX 3090. Lower-VRAM cards may work if you reduce resolution or batch size, but expect slower runs.
Can I run this without paying per image?
Yes, generation runs locally in ComfyUI, so there are no hosted per-image charges. You still need the hardware and model access.
This guide was drafted from the video's details and general task knowledge. It has not been checked against the full video. Check the video and relevant product instructions before following a procedure. Report a correction.
