RedNode Studio (Krea 2)

Start here. That's not a suggestion from me - it's the pack's own framing, and it's right. Krea2RedNode wraps the moodboard and identity-edit machinery behind one simple node with a matched grounded negative built in, which removes the single most common way people break identity-edit workflows: wiring the positive and negative up wrong.

What it gives you

Two outputs, positive and negative, both CONDITIONING. The negative isn't an empty text encode - it's a properly matched grounded negative, built from the same sources, the same VAE, the same fit geometry as the positive. In the legacy path you assemble that by hand with two Krea2IdentityEdit nodes and hope you connected everything consistently. Here it's one node, and the mismatch class of bugs just disappears.

The inputs you'll actually set

• instruction - the edit instruction, e.g. "create a photo of this person at a night market." Also doubles as the prompt for pure style transfer.
• preset - four modes, and they cover the realistic use cases:
  • balanced - identity + style fusion.
  • max identity - strongest face lock.
  • style only - pure vibe transfer, no LoRA needed at all.
  • outfit transfer - puts the clothes from moodboard_style onto the subject (style_strength 1.0; for strict garment construction also feed the garment into scene_image and drop style_strength to 0.8).
• subject_image - the face/person to preserve.
• scene_image - optional environment reference for two-ref edit LoRAs. Ordering (scene first, subject second) is handled for you, which is another thing the legacy nodes make you manage.
• moodboard_style - your style references; batch several for a joint moodboard.
• style_strength - how much of the style survives (0.5 default; 1.0 = raw reference detail).
• vae - required the moment a subject or scene image is connected.
• output_latent - connect the same empty latent you feed the sampler. This enables the v1.2 blur-proof fit geometry on both outputs at once, and it's the one connection people skip. Don't skip it.
• extra_subjects - chained references from Krea2EditSourceChain (3+ blends identities, same caveat as the rest of the pack).
• settings - plug a Krea2RedNodeSettings node in here and it replaces the preset entirely, giving you every dial in plain language.

Wiring

clip → Krea2RedNode → positive → KSampler positive
                       negative → KSampler negative
         output_latent ← your empty latent (same one the sampler uses)

That's the whole integration. The pack ships two lean example workflows that run on core nodes plus this pack alone - krea2_rednode_identity.json (max identity preset) and krea2_rednode_style_transfer.json (style only, no LoRA). The full advanced pipeline (Krea_Workflow_Public.json) exists too, but it pulls in a dozen other packs; you don't need any of them to get going.

Notes on the mechanics

The node is the same dual-conditioning recipe as Krea2IdentityEdit - in-context identity latents at RoPE frames plus Qwen3-VL-grounded instruction - with the moodboard fused in. The positive/negative pairing is the training unconditional, so it's correct by construction here. With the v1.2 edit LoRA, run 8–12 steps on Turbo (8 favors composition, 12 favors face detail), and generate at ≤2MP for best results.

Installing

Same as the whole pack - GPL-3.0, no third-party node dependencies:

git clone https://github.com/RedNodeAI/ComfyUI-Krea2Moodboard ComfyUI/custom_nodes/ComfyUI-Krea2Moodboard

or search "Krea2Moodboard" in ComfyUI Manager, then restart. You need ComfyUI 0.25.0+, qwen3vl_4b (CLIPLoader type krea2), qwen_image_vae, and a krea2_edit LoRA at strength 1.0 for any preset that touches identity. If you're new to this pack, this is the node to learn first - everything else is the advanced path underneath it.