Krea 2 Moodboard + Identity Edit - ComfyUI Nodes (+ Forge Neo) — v1.3 — Rednode Studio 1.6.4

RedNode Studio 1.6.4

(everything since 1.4.3) New: a full video tutorial. Install, models, and every tab you need to make, fix and save a picture, from a blank canvas to a finished render in 30 seconds. https://www.youtube.com/watch?v=GSwWz7awvpw

Update through ComfyUI Manager, restart, and load the RedNode Studio V1.6 template.

On Windows, the two .bat files on the GitHub release install the node packs and download the models for you.

The big changes

• A new tab rail down the side, grouped with icons. Fold it, resize it, put it on either side, and pick a tab preset for the job: Basic, New image, Edit, Paint and more.

• Paint: AnyPaint redraws just the area you paint with the krea2_anypaint LoRA so it fits the scene, about 8 seconds a paint. Plus Re-render as a Paint engine and a history with full screen viewing.

• Re-render: turn an illustration into a photograph, as its own tab, a Detailer pass or a Paint engine, with a Repeat count.

• Tools: Upscale (SeedVR2, VOSR 2.0 or tiled, one picture or a folder), Re-angle, Swap and the Converter.

• Hero Creator: a clean front-on headshot out of any photo, for the Identity tab.

• RedNode Shelf: drop pictures on it and drag them anywhere, with five stacks and an Override that makes the picked picture the one a tab renders from.

• AI tab: pictures to words on a page of its own, and Image to Text as a Detailer pass.

• Run tab: Live, Review and Save, with this session's runs down the side. Models - Faster Krea 2 by default: the workflow and the download use Comfy-Org's official INT8 Krea 2 Turbo, half the VRAM of the full model, close to it in quality, and quicker. Needs ComfyUI 0.27 or newer. Your current Krea 2 file still works, just pick it in the rig.

• Qwen Image 2.1 as a rig, with Size Shift. - Models tab pages: Model, CLIP, VAE, Sampling, Seed and Setup. Setup builds a rig in one click for Krea 2, Z-Image, Qwen Image 2.1 or SDXL / Illustrious / Pony.

• CLIP Skip on any rig's text encoder.

• Plain Text Encode on a Krea 2 rig, to match a plain workflow seed for seed.

• Rigs can sample through RES4LYF's sampler. Sampling

• Three sampler presets ship with the pack, picked from side-by-side tests on Krea 2: Krea 2 Crisp, Krea 2 Crisp Detail and Krea 2 Soft. Presets now save the Sampler dials too.

• Exact LoRAs: LoRAs run at full precision instead of being baked into fp8 or INT8 weights, so sliders at low strengths keep their effect.

• Focus, and a Look Only mode for Detail Daemon that sharpens Krea 2, Z-Image and Flux instead of softening them.

• The Passes page draws the noise schedule your run will use, and the rig card warns about sampler pairs that burn Krea 2. Prompts and LoRAs

• Importing a picture from A1111, Forge or Neo reads its LoRA tags into the LoRA stack.

• Prompts render on whichever rig is active, and each prompt can pick its own LoRA set.

• Every LoRA set has its own switch, and the stack has a search box. Important fixes

• The Moodboard works properly again. With the Studio preset on custom, the Moodboard and Identity dials were ignored and every run fell back to Balanced. If the Moodboard felt like it did nothing, this was why.

• Re-render no longer fills the graphics card run after run. - Prompts imported from a picture no longer render empty.

Model needed for AnyPaint: krea2_anypaint ( huggingface.co/yijunwang2/krea2-anypaint ), into models/loras.