Krea2 Txt2img

Krea 2 Text-to-Image Workflows

Overview

Krea 2 is a 12B-parameter Diffusion Transformer from Krea.ai (released June 2026, weights open-sourced under the Krea 2 Community License, free commercial use up to 50 seats). Two variants:

1. Krea 2 Raw is the base checkpoint before extra post-training. For fine-tuning / maximum fidelity, more steps.
2. Krea 2 Turbo is post-trained and distilled; it generates in ~8 steps at cfg 1. This is what the krea2 txt2img packs ship.

Three packs (V2 — no group toggles)

Sliced from the KREA2 ULTRA V2 monolith into standalone single-pipeline packs. Pick by how you prompt and what you want:

• krea2-txt2img-manual: plain prose prompt (the MANUAL PROMPT node).
• krea2-txt2img-json: Ideogram-4-style structured JSON / area prompting (Ideogram4PromptBuilderKJ).
• krea2-combo: two-pass detail boost, a first pass then a low-denoise refine (denoise 0.3), with the krea2 turbo LoRA @0.2 on both passes plus the optional IdeoKrea LoRA. JSON/Ideogram-style prompting; saves both passes to compare.

Each pack's one prompt source is active (no prompt-mode bypass to flip). ImageSharpenKJ runs before SaveImage. V2 adds the Krea2T-Enhancer MODEL detail-boost patch (ships active) and drops v1's ConditioningKrea2Rebalance. RBG_Smart_Seed_Variance ships bypassed (optional, see below).

Krea 2 has native ComfyUI support (comfy/text_encoders/krea2.py, ComfyUI ≥ v0.26.0). The CLIPLoader uses type=krea2, with a Qwen3-VL 4B text encoder and the Qwen image VAE. The Qwen3-VL encoder drives strong prompt adherence and structured-JSON prompts.

Models (all from the Aitrepreneur/FLX mirror; official: krea/Krea-2-Turbo)

Node stack

• core: UNETLoader (krea2_turbo) → CLIPLoader (type=krea2) → VAELoader (qwen_image_vae), wired via KJNodes SetNode/GetNode buses into a subgraph (CLIPTextEncode → KSampler → VAEDecode). An rgthree Any Switch sits in front of the encoder; in each pack only that pack's prompt source is wired to it (manual node in -manual, JSON builder in -json).
• rgthree-comfy: Power Lora Loader, Any Switch, Label, Fast Groups.
• ComfyUI-KJNodes: Set/Get, Ideogram4PromptBuilderKJ, ImageSharpenKJ, INTConstant.
• ComfyUI-Krea2T-Enhancer (capitan01R): Krea2T-Enhancer, the V2 MODEL→MODEL detail-boost patch, wired in the model path (PowerLora → Krea2T-Enhancer → sampler). Ships active; bypass to compare against the un-boosted result.
• ComfyUI-RBG-SmartSeedVariance: RBG_Smart_Seed_Variance, optional, ships bypassed in the positive-conditioning loop.
• ComfyUI_essentials (cubiq): ImageResize+, combo only (the two-pass VAE-roundtrip resize).

Settings that matter

• steps 8, cfg 1. Turbo is distilled; more steps or higher cfg over-cooks it.
• sampler er_sde, scheduler simple are the verified defaults.
• 1920×1080 default; Krea 2 handles a wide aspect range.
• The prompt source is fixed per pack (manual node vs JSON builder). There is no prompt-mode bypass to flip.

V2 detail boost (Krea2T-Enhancer) + combo

• Krea2T-Enhancer is a MODEL→MODEL patch (the V2 "massive detail boost"). It sits inline in the model path and ships active in all three packs. Widgets are [on, strength, …]; bypass it (or toggle on) to A/B the boost.
• krea2-combo is the full demonstration of the boost, a two-pass refine: FIRST PASS (8 steps, er_sde, denoise 1) → VAE roundtrip → SECOND PASS (4 steps, euler, denoise 0.3), with the turbo LoRA @0.2 on both passes. It SAVES BOTH passes so you can see the boost. The IdeoKrea LoRA is downloaded but NOT wired by default. Drop it into the Power Lora Loader's empty slot (start ~0.5 to 1.0; it's a test LoRA) for the turbo + IdeoKrea Ideogram-style combo.