A two-pass workflow for Jib Mix Moss Realistic by J1B, my favourite MOSS finetune after sweeping eight of them with my LoRA. Thanks, Jib, for a model that keeps a character's face while making the photo look real. The full comparison is in my article Which base model is best for your LoRA?
What it does
1st pass: 1024×1536, 12 steps, CFG 1, euler / simple, stopped at step 10 so the latent keeps its leftover noise. That's Jib's "recommended 12 steps" (the 8-step LoRA is merged into the model).
1.5× latent upscale, then a 2nd pass on the same 12-step schedule from step 3, which finishes at 1536×2304.
SeedVR2 7B up to 4096 px, blended with 4x-ClearRealityV1 Soft (on). RTX Upscale is optional and off by default.
LoRA Manager loader with the mbl9 LoRA at 1.0. Its trigger words are added in front of the prompt automatically, through LoRA Manager's TriggerWord Toggle.
Prompt Manager text node (every prompt is saved to its library), Kiko Prompt Enhancer (off): rewrites a short prompt into a full MOSS caption with your local text encoder and keeps the trigger words exact.
One seed drives both passes, and Preview 1st pass shows the composition early.
The two-pass technique (stop the 1st pass early with leftover noise, upscale the latent, finish on a fresh schedule) comes from catlover1937's Moody Moss Mix workflow. I scaled it from his 18/25 steps to Jib's 12.
Tuning (in the note): a 2nd-pass start step of 2 reworks more, 4 keeps the 1st pass layout. Jib himself runs er_sde / bong_tangent at 4–9 steps. bong_tangent needs RES4LYF; er_sde with beta is the closest built-in pair.
Files
kiko-jib-mix-moss-v7.json: the workflow.kiko-jib-mix-moss-workflow.png: a picture of the whole graph, with the workflow embedded.Example renders: mbl9 at 1.0, three seeds, the default prompt.
Custom nodes
ComfyUI-KikoTools (1.6.0+)
ComfyUI_essentials (SimpleMath+)
Optional: ComfyUI_Nvidia_VFX_Nodes, for the RTX Upscale group (NVIDIA RTX only)
Installing KikoTools: the Comfy Registry lists KikoTools as flagged, so ComfyUI Manager 4.2 (the version ComfyUI ships) won't install it. Either update the Manager to 4.3 or newer and set allow_flagged_nodepack_install = true in ComfyUI/user/__manager/config.ini, or install from GitHub: git clone https://github.com/ComfyAssets/ComfyUI-KikoTools in ComfyUI/custom_nodes, then pip install -r requirements.txt with ComfyUI's Python, and restart.
Models
The bypassed Download models group fetches all of them. Turn it on in Toggle stages, queue once, then turn it off again.
Moss/Jib_Mix_Moss_V3_B_00001_.safetensors→models/diffusion_models/: Jib Mix Moss V3 (bf16. int8 and fp8 files for smaller cards are on the same page)qwen3.5_9b_w4a8_gptq.safetensors→models/text_encoders/: https://huggingface.co/Comfy-Org/MOSS-Image/resolve/main/text_encoders/qwen3.5_9b_w4a8_gptq.safetensorsFlux.2-D/flux2-vae.safetensors→models/vae/: https://huggingface.co/black-forest-labs/flux-3-image-klein-comfy/resolve/main/vae/flux2-vae.safetensors4x-ClearRealityV1_Soft.safetensors→models/upscale_models/: https://huggingface.co/Kim2091/ClearRealityV1SeedVR2's models download themselves on first use.
mbl9 LoRA →
models/loras/
Civitai downloads need your API key, set where ComfyUI starts, never in the workflow: CIVITAI_API_KEY (Linux/macOS export, Windows setx). Then restart ComfyUI.
Happy rendering! Kiko
Description
First public release. 12 steps, CFG 1, euler / simple, two passes plus SeedVR2. Set up for the mbl9 LoRA with the Comfy-Org qwen3.5_9b_w4a8_gptq text encoder; the Download models group fetches every model, the LoRA included.



