Krea-2 Turbo · Dual-Mode QwenVL Workflow — Keyword Prompt or Image Reference, No Rewiring · + Inline SeedVR2 Upscale — v1.0

Author notes from krea2TurboDualModeQwenvlWorkflow_v10.json · selected release

⚖️ LICENSE Krea-2 model: krea-2-community-license ✓ Commercial use allowed ⚠️ Content filtering required — no NSFW Weights (UNet + CLIP + VAE): huggingface.co/AlperKTS/Krea2_FP8 QwenVL: Apache-2.0 (1038lab/ComfyUI-QwenVL) LLM auto-downloads: Qwen/Qwen3-VL-2B-Instruct Workflow: CC0 — free to use, share, modify

Author notes from krea2TurboDualModeQwenvlWorkflow_v10.json · selected release

📸 MODE B — IMAGE REFERENCE → AI DESCRIBE (inactive by default, index=1) Load any reference image below → QwenVL reads & describes it → Krea-2 generates inspired by its style and content. To activate: set 🔀 Mode Switch [index] = 1 | To return to Mode A: set [index] = 0

Author notes from krea2TurboDualModeQwenvlWorkflow_v10.json · selected release

✍️ MODE A — SUBJECT KEYWORD → AI PROMPT (active by default, index=0) Type any subject keyword or phrase in [prompt_text] below. Qwen3-VL-2B (local, ~3 GB, auto-downloads on first run) expands it into a rich, photorealistic prompt for Krea-2 Turbo.

Author notes from krea2TurboDualModeQwenvlWorkflow_v10.json · selected release

🔀 MODE SWITCH (EasyUse Index Switch) index = 0 → Mode A ✍️ PromptEnhancer (default) index = 1 → Mode B 🔍 QwenVL VL (image reference) Change [index] widget to switch. No rewiring, no bypass — just a number.

Author notes from krea2TurboDualModeQwenvlWorkflow_v10.json · selected release

📐 LATENT · index 0 = Landscape 1920×1088 · index 1 = Portrait 1088×1920

Author notes from krea2TurboDualModeQwenvlWorkflow_v10.json · selected release

🚀 QUICK START ▸ MODE A (keyword → AI-expanded prompt) 1. Type subject in ✍️ PromptEnhancer [prompt_text] 2. Click Queue Prompt ▸ SWITCH TO MODE B (image reference) 1. Set 🔀 Mode Switch [index] = 1 2. Drop image into 📸 LoadImage 3. Click Queue Prompt ↩ Back to Mode A: set [index] = 0 ───────────────────────────────────────────── ▸ ASPECT RATIO 📐 Latent Switch [index]: 0 = Landscape 1920 × 1088 (16:9) 1 = Portrait 1088 × 1920 (9:16) ───────────────────────────────────────────── Both switches: just change a number. No rewiring needed. ───────────────────────────────────────────── ▸ MODELS (download before first run) Manual → place in ComfyUI/models/ unet/ krea2_turbo_fp8.safetensors clip/ qwen3vl_4b_fp8_scaled.safetensors vae/ qwen_image_vae.safetensors Source: huggingface.co/AlperKTS/Krea2_FP8 Auto-download on first Queue (~4 GB): Qwen3-VL-2B-Instruct (via QwenVL node) ───────────────────────────────────────────── Sampler: euler · simple · 10 steps · cfg 1.0 Output: ComfyUI/output/krea2_turbo/