LIVE WIRE Ubuntu vs Windows 11 for AI Image Generation in 2026: A Careful ComfyUI Comparison
[ SYSTEM DISPATCH // V.4.2 ]
AI INTELLIGENCE & GENERATIVE SYSTEMS
[ FILED UNDER // COMFYUI ]

MiniMax H3 Character Sheet V2: Build, Dress, and Reshape One Consistent Character

AIwithWalt October 22, 2026 5 min read

This video introduces Version 2 of MuseCollective’s MiniMax H3 Character Sheet workflow for ComfyUI. It is a major upgrade over V1, adding multi-reference mixing, outfit and body-shape control, and customizable sheet templates. The workflow itself is exclusive to the Premium / Advanced Tier of his Skool Community.

From V1 to V2: What’s New

The creator frames V2 as a more advanced, more customizable version of the original H3 character sheet node. The core idea is unchanged: generate a video turnaround from a reference image, then crop frames into a printable character sheet.

What is new is flexibility. V2 adds different sheet modes, reference roles for combining images, SAM segmentation for clothing and props, a workspace tool for framing panels, and optional Qwen 2.1 edit + DLSS-style enhancement passes. For speed in the demo he runs at 0.4 megapixels with single-stage sampling, using Singularity H3 INT8 with normal VAEs and Kijai’s experimental 4-step DMAD LoRA.

Still from MiniMax H3 Character Sheet - V2
Still from “MiniMax H3 Character Sheet – V2”.

Sheet Modes and Basic Turnaround

V2 offers five modes: Basic Turnaround, Basic + Expressions, Basic + Detail, Basic + Angles, and Complete Capture. Selecting a mode beyond Basic appends extra prompt chunks, renders them separately, then joins everything into one long output video.

The demo starts with a single female reference. After clicking Analyze to generate a description, he runs a Basic sheet with trigger words off because no realism LoRAs are attached. The result is a full 360 turnaround plus a close-up. Frames are then placed into a template: close-up for the hero panel, front, 3/4, side, and back views for the others, using Fill, zoom slider, and pan to frame each shot. From there you can download the PNG directly, or Continue to Output to enhance it through Qwen.

Still from MiniMax H3 Character Sheet - V2
Still from “MiniMax H3 Character Sheet – V2”.

Mixing References: Outfits, Body Shape, Attributes

The core of V2 is reference roles: Main Character, Clothing, Accessories, Handheld Items, Body Shape, Style Transfer, and Attribute Transfer. This lets you combine multiple images into one consistent character.

First he adds a gym outfit — black sports bra and leggings — set as Clothing. SAM segmentation isolates the clothing so the identity does not bleed over from the outfit model. After re-running, the same woman now wears the gym outfit, watch included.

Next he tests body shape with an extreme muscular reference set to Body Shape. The result is noticeably more muscular with visible abs, while face and identity stay largely consistent. He notes putting body shape second seemed to work better in testing, and that results can vary by model and LoRA.

He then stacks attribute transfers: deep red lipstick from a lips close-up, and later pigtails in a Harley Quinn style. Because the hair description mentions placement and parting but not color, only the style transfers. He stresses editing the auto-descriptions if they contain unwanted details — for example removing mention of crystals on a neckline.

Still from MiniMax H3 Character Sheet - V2
Still from “MiniMax H3 Character Sheet – V2”.

Props, Hair, and Stacking Limits

Handheld items get a left/right hand selector. A water bottle in the right hand works on the first try, giving a final character with custom outfit, muscles, red lips, pigtails, and bottle in hand.

But stacking has costs. When red lips were added, an unwanted sweatshirt appeared; a new seed fixed it. The workflow does not yet have the retake system from his V2 Director workflow, so the workaround is to change the seed and re-run.

A second character test — new face, shiny outfit, different hair, red lips, and a samurai sword for the left hand — shows the same trade-off. Wings described via ChatGPT assistance do appear but look cartoony, and the sword drops out on that seed. A later Complete Capture run does bring the sword back. His advice: keep prop descriptions short like “samurai sword” and disable SAM if the verbose description hurts.

Still from MiniMax H3 Character Sheet - V2
Still from “MiniMax H3 Character Sheet – V2”.

Templates, Framing, and Complete Capture

Finished videos populate the template area at the top. You can switch resolution, keep the video background or replace it with white or light grey, choose from five templates, then Fill / Crop, zoom, and pan each panel before hitting Update Sheet Preview.

Complete Capture is the showcase mode: 1) turnaround, 2) expressions — scared, happy, angry, surprised, 3) detail crops like right hand, belt buckle, and eyes, and 4) camera angles including top-down view. It produces one long video you scrub to find frames. Even at 0.4 MP, close-ups hold up well; full-body panels need higher resolution. He suggests blacking out heads on body panels is common practice, since the close-ups already cover face and expression.

He closes by showing the upscaler path: 0.98 MP first pass upscaled to 2K on the second pass, still at four steps, yielding photographic skin detail, sharper leather, and catchlights and freckles in the eyes.

Key takeaways

  • V2 turns one image into a video turnaround + close-up, then into a framed PNG character sheet with 5 templates.
  • Six reference roles let you mix identity, clothing, body shape, accessories, handheld props, and fine attributes like lipstick or hairstyle.
  • SAM segmentation is key for clothing and props to avoid identity mixing, and auto-descriptions should be manually cleaned.
  • Stacking many changes works but drifts — expect to re-seed; no integrated retake yet.
  • Complete Capture adds expressions, detail inserts, and angles in one long render for richer sheets.
  • Run as high resolution as your GPU allows, and use Qwen + upscale for final quality; 0.4 MP is only for fast testing.

Who it is for

This is for ComfyUI users building consistent original characters for illustration, comics, games, or AI video who already have MiniMax H3 working and want outfit variants, physique changes, and expression sheets without losing face identity. Free Skool members can watch for ideas, but to use this exact V2 workflow you need the Premium / Advanced Tier.

Watch it

You can watch the full 51-minute walkthrough here: MiniMax H3 Character Sheet – V2 by MuseCollective.

Leave a Reply

Your email address will not be published. Required fields are marked *

>_ JOIN THE INTELLIGENCE NETWORK

One email when a new breakdown drops: the workflow, the settings, and what actually worked. No spam, unsubscribe anytime.

Join the Intelligence Network

New AI video & image breakdowns, straight to your inbox. Free, no spam.