Workflow is able to +4K Text-to-Image and also Multi-Reference Image-to-Image.UPDATES: v3.0 - Removed SDE node and replaced with 3 new nodes. Removing SDE and adding the others gives a much more detailed natural look to skin, while a bit softer the tradeoff was worth it IMO, also seems to have better prompt adherence. v2.0 on left, v3.0 on right<img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/b02817c7-b758-4216-9957-9374dff5ed96/original=true/b02817c7-b758-4216-9957-9374dff5ed96.jpeg" />Clown-Shark Workflow v3.0:<img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/47af5c44-8090-4e6d-a2e5-994d5fbae32b/original=true/47af5c44-8090-4e6d-a2e5-994d5fbae32b.jpeg" />It'’s got two sampler groups: the regular Sampler Custom, and the Clown Shark Sampler from RES4LYF, you can swap between the working groups with the Fast Groups Muter.Plus several of its optional nodes — Detail Boost, Implicit Steps, Momentum and Sigma Scaling. I left notes from the official RES4LYF example workflow for explanations of what they do, along with some quick notes on my testing of them.I tried most of the samplers and landed on what feels like the sweet spot for Clown Shark: diag_implicit/pareschi_russo_2s with Beta57. It’s a little slower, but totally worth it IMO.The Klein node for prompting (in System Settings Subgraph) you can just delete from the workflow or disable if you don't want it, but I find it does help.Quick notes:<ul><li>The DasiwaMinimaxH3_dasiwaREF2VAHybridV1 Darksidewalker pulled is a must, every setting the workflow is tuned for this exact model, if you use anything else you will get lackluster results. I tried them all, they are all oversaturated with a plastic look to them.</li><li>I'd say this is a tweakers workflow, I you don't like hunting down that last setting and just want a set it and forget it workflow, this one isn't for you.</li><li></li><li>The Beta57 scheduler by RES4LYF is a must for image quality in this workflow.</li></ul><ul><li>For extra detail, run a SeedVR detail pass with the same resolution as the image (no upscaling).</li><li>I’d keep the LoRA strength light as these were made for video.</li><li>Eyes are a problem when distance is involved and prompt adherence spotty, let's hope the image model fixes it.</li><li>If you use JonXL's Generator LORA with text-to-image, I found a strength setting of .20 @8 steps is best, with reference images you can go higher.</li><li>The model I used to create the images was<a target="_blank" rel="ugc" href="https://civarchive.com/models/2877206/dasiwa-minimax-h3?modelVersionId=3272675"> </a>DaSiWa MiniMax H3 and the CLIP was Huihui-Qwen3-8B-abliterated-v2-FP8-Comfy as these both gave much better results.</li><li>Pretty much everything you can tweak is sitting on the subgraphs. Fair warning: cranking the resolution too high will probably throw errors.</li><li>My pin naming is a bit weird, but that’s on purpose. It helps me keep track of what goes where when I’m unplugging and plugging a bunch of stuff.</li><li>I put the Save Image node on a switch, I only save the ones I actually like. Leave the seed on fixed, generate, then flip the switch and hit run to save (or just right-click the preview and save it manually).</li><li>Reference Image input size has a great deal to do with rendering time, the bigger the image the slower the renders. I set the resize @2.0, if you have a slower card set this lower.</li></ul>Nodes and models you will need to have installed, the list below is the bare minimum.custom_nodes:<ul><li><a target="_blank" rel="ugc" href="https://github.com/cicalooo/ComfyUI-MM-1Frame">ComfyUI-MM-1Frame</a></li><li><a target="_blank" rel="ugc" href="https://github.com/yolain/ComfyUI-Easy-Use">ComfyUI-Easy-Use</a></li><li><a target="_blank" rel="ugc" href="https://github.com/capitan01R/ComfyUI-Flux2Klein-Enhancer">ComfyUI-Flux2Klein-Enhancer</a></li><li><a target="_blank" rel="ugc" href="https://github.com/kijai/ComfyUI-KJNodes">ComfyUI-KJNodes</a></li><li><a target="_blank" rel="ugc" href="https://github.com/ClownsharkBatwing/RES4LYF">RES4LYF</a></li><li><a target="_blank" rel="ugc" href="https://github.com/rgthree/rgthree-comfy">rgthree-comfy</a></li></ul>## Model Linksdiffusion_models:<ul><li><a target="_blank" rel="ugc" href="https://huggingface.co/Comfy-Org/MiniMax-H3/resolve/main/diffusion_models/minimax_h3_ref2va_pruned_int8_convrot.safetensors">minimax_h3_ref2va_pruned_int8_convrot.safetensors</a></li></ul>text_encoders:<ul><li><a target="_blank" rel="ugc" href="https://huggingface.co/Comfy-Org/MiniMax-H3/resolve/main/text_encoders/qwen3vl_32b_minimax_h3_nvfp4_awq.safetensors">qwen3vl_32b_minimax_h3_nvfp4_awq.safetensors</a></li></ul>vae (HARD REQUIREMENT):<ul><li><a target="_blank" rel="ugc" href="https://huggingface.co/Mamad8/MiniMax-H3-Image-VAE/resolve/main/minimax_h3_t1_image_vae_step1597.safetensors">minimax_h3_t1_image_vae_step1597.safetensors</a></li></ul>vae_approx (TAE):<ul><li><a target="_blank" rel="ugc" href="https://huggingface.co/Kijai/MiniMax-H3-TAE/resolve/main/vae_approx/taeh3.safetensors?download=true">taeh3.safetensors (MiniMax H3 TAE)</a></li></ul>loras/h3:<ul><li><a target="_blank" rel="ugc" href="https://huggingface.co/Kijai/MiniMax-H3_comfy/resolve/main/loras/minimax_h3_fl2v_lightx2v_turbo_8step_v1.0_resized_avg_rank_24_bf16.safetensors">minimax_h3_fl2v_lightx2v_turbo_8step_v1.0_resized_avg_rank_24_bf16.safetensors</a></li><li><a target="_blank" rel="ugc" href="https://huggingface.co/Mamad8/MaxiMin-HHH-R2V-ThisIsFine/resolve/main/MaxiMin-HHH-R2V-ThisIsFine_LoRA_V0_1.safetensors">MaxiMin-HHH-R2V-ThisIsFine_LoRA_V0_1.safetensors</a> (Optional)</li><li><a target="_blank" rel="ugc" href="https://civarchive.com/models/2884206/minimax-h3-photorealistic-image-generator-lora-workflow">Minimax H3 Photorealistic Image Generator LoRA</a> (Optional)</li></ul>## Highly Recommendeddiffusion_models:<ul><li><a target="_blank" rel="ugc" href="https://civarchive.com/models/2877206/dasiwa-minimax-h3">DaSiWa MiniMax H3</a> (REF2VA Turbo or Non-Turbo)</li></ul>text_encoders:<ul><li><a target="_blank" rel="ugc" href="https://huggingface.co/YifeiDevs/Huihui-Qwen3-8B-abliterated-v2-FP8-Comfy/resolve/main/Huihui-Qwen3-8B-abliterated-v2-FP8.safetensors?download=true">Huihui-Qwen3-8B-abliterated-v2-FP8-Comfy</a></li></ul><img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/4454e2b0-15a2-429c-aeff-0895f37d9ff4/original=true/4454e2b0-15a2-429c-aeff-0895f37d9ff4.jpeg" />Example Prompts:<pre><code>Single-frame photoreal still. Dew-field macro.

A praying mantis nymph the color of new barley, perched on a seed head, surrounded by a galaxy of dew. Each drop holds an inverted green world. The mantis’s raptorial forelegs folded, head turned a few degrees toward camera.

Color science: young-wheat green, silver dew, a hint of chartreuse in the eyes. Backlit dawn so every drop becomes a lens.

100mm macro, focus on the mantis head, dew field glittering out. Photoreal, no text.</code></pre><pre><code>Single-frame photoreal still. Breasts-up, camera slightly above eye level.

Mid-30s Italian woman in a shadowed trattoria doorway in Lecce. Olive skin, dark wavy hair over one shoulder, chin down, eyes up. Soft hourglass frame, medium teardrop breasts.

Color science: oxblood–cream–old-gold. Cream ribbed tank under an open oxblood cardigan that frames the chest. Makeup: apricot blush, bronze-olive lid, terracotta-nude satin lip, espresso brow. Gold hoops pick up the trattoria brass.

Location: flaking apricot plaster, dark interior behind her, one warm pendant bulb as a small highlight. Natural shade with warm bounce.

100mm, photoreal, no text.</code></pre><pre><code>Single-frame photoreal still. River-valley panoramic at blue hour.

A wide boreal oxbow in Finland. A pair of whooper swans on black water, white shapes doubled in the reflection. Bog cotton and dark spruce on the far bank, the last strip of peach sky under a lid of indigo.

Color science: indigo, peach, swan-white, spruce-black. Still water, no ripple except one swan’s wake.

35mm, low camera just above the reeds, huge quiet negative space of sky and water. Photoreal, no text, no people.</code></pre><pre><code>Cinematic dawn photograph of a beautiful woman in her mid-20s standing in sea mist on a coastal pull-off, unzipping a battered leather jacket over a black dress she rode in, helmet in one hand, the other on the seat of her BMW R nineT. 
High cheekbones, dark wet lashes, the kind of beauty that photographs well and still looks cold and tired.

The R nineT is accurate: boxer cylinders, classic tank, correct tail, spoked or aluminum wheels as spec, salt rime on the engine cases, scuffed bar-end.

Color science: 6000K blue dawn lifting to 4500K sun, Portra 160, pale cool skin with a warm kiss on the nose and forehead, BMW black and silver kept honest, sea-green mist, soft contrast.

She is stunning and she got here on that bike.</code></pre><pre><code>Use Image 1 as the exact character reference. Keep her exact face, identity, long straight black hair with blunt bangs, pale skin, eye color, body type, and proportions. Do not change her hairstyle or body.

Dress her in the outfit from Image 2: navy textured blazer, black satin camisole with lace trim, charcoal pinstripe high-waisted wide-leg trousers, black belt with gold buckle. Makeup from Image 2: porcelain ivory skin, matte obsidian lipstick, smokey ash eyeshadow, soft cool mauve blush.

Place her in the room from Image 3, seated at the rustic wooden table by the sunlit window. She is taking a quiet break from writing. A steaming coffee cup sits on the saucer in front of her.

The open ledger shows only handwritten pages. The pages are empty of objects. She holds the fountain pen loosely in her right hand, lifted off the book. The table surface around the book is clear except for the coffee cup and saucer.

Keep the warm window light, wood texture, steam, and atmosphere from Image 3. Photorealistic, cinematic, highly detailed, 9:16 vertical composition.</code></pre>

MiniMax H3 - Text-to-Single-Image | Multi-Reference Image-to-Image Workflow

<ul><li>Initial release - v1.0</li><li>v2.0 - Added multi-reference image loading, added on/off switches for Detailer and SDE nodes, minor wording fixes. Ended and removed Basic workflow version, only releasing Clown-Shark versions moving forward.</li><li>v2.0 ReUpload - Cleaned up some of the mess I left, added some notes with node and model links.</li></ul>

Files with SHA256: a44a32dd51dd94b0ca9f726248e4c5b43b6d30d46310c343e4f7cf2661851ec8