⚗ AlchemicaSampler for Krea 2 TurboA sampler that measures your checkpoint, then samples the way the model actually works.⚗ Idea and direction: AlchemicaMente · code: Claude (Anthropic) · license: GPL-3.0 Custom nodes and full documentation: <a target="_blank" rel="ugc" href="https://github.com/Nessuno76/ComfyUI-AlchemicaSampler">github.com/Nessuno76/ComfyUI-AlchemicaSampler</a><h3 id="what-this-workflow-does">What this workflow does</h3><ol><li>Prompt enhancer built on Krea 2's own text encoder. Write an idea in any language (or connect a reference image). Qwen3-VL-4B, the very encoder Krea 2 uses, writes a SCENE, a DETAIL and a NEGATIVE prompt. No Ollama, no external server.</li><li>NAG, a negative prompt that works at cfg 1. Krea 2 Turbo is distilled, so the usual negative prompt does nothing. NAG applies the negative inside the attention of every DiT block. Starting at the calibrated hand-off sigma, it removes what you don't want without changing the picture (measured: 99.3 % identical image, unwanted water drops gone).</li><li>Two-phase prompt. The scene leads until the composition locks (σ ≈ 0.87, measured); the detail prompt leads after that. Same picture, different rendering of skin, fabric and materials.</li><li>Calibrated sampling. A calibration run measures your checkpoint on your GPU and builds the step schedule, the two-stage hand-off and the detail window from those measurements.</li><li>Composition previews. Generate 4–8 small previews, pick one, render it at 4 MP with the same composition.</li></ol><h3 id="requirements">Requirements</h3><ul><li>ComfyUI 0.36.0+</li><li>Custom nodes: ComfyUI-AlchemicaSampler (GitHub link above, or search "AlchemicaSampler" in ComfyUI-Manager)</li><li><code>krea2_turbo_nvfp4.safetensors</code> or <code>krea2_turbo_fp8_scaled.safetensors</code> → <code>models/diffusion_models</code></li><li><code>qwen3vl_4b_fp8_scaled.safetensors</code> → <code>models/text_encoders</code></li><li><code>qwen_image_vae.safetensors</code> → <code>models/vae</code></li><li>16 GB VRAM for 4 MP (NVFP4). On RTX 50-series use a PyTorch build for CUDA 13.0 (cu130) for full NVFP4 speed.</li></ul><h3 id="how-to-use">How to use</h3><ol><li>Install the custom nodes and restart ComfyUI.</li><li>Run KREA2_TARATURA once with your checkpoint (≈15–20 min at 4 MP). It is optional but recommended: without it the nodes fall back to default values.</li><li>Open KREA2_COMPLETO, type your idea in the enhancer, press Queue.</li><li>To remove something from the image, write it in the negative (or keep the enhancer's). NAG 3 from −1 is preset.</li><li>To explore compositions: set <code>anteprime</code> (previews) = 8 with the seed on fixed, pick a number, set <code>anteprime</code> = 0 and <code>scelta</code> (choice) = that number.</li></ol><h3 id="recommended-settings-(validated)">Recommended settings (validated)</h3><ul><li>Resolution: 4 MP (5 MP only for close-ups)</li><li>modo (mode): equilibrato (balanced)</li><li>dettaglio (detail boost): 0.12</li><li>varieta (variety): 0.6</li><li>NAG: scale 3 · tau 2.5 · alpha 0.25 · from −1</li><li>Structure guidance: off (PAG 0.5–1 from −1 for a crisper, more contrasty look)</li></ul>Timing on an RTX 5070 Ti at 4 MP: ~63 s without NAG, ~85–95 s with NAG. The first run of the complete workflow takes longer because the text encoder writes the prompts; they stay cached until you change the idea.<h3 id="honest-notes">Honest notes</h3><ul><li>The detail prompt changes texture, not colours or style: Krea 2 Turbo decides those in the first step.</li><li>PAG is a style (crisper, fewer freckles), not an absolute improvement. SEG above 0.3 makes halos.</li><li>NAG and the enhancer are written for Krea 2 only.</li></ul><h3 id="license">License</h3>Workflows and nodes: GPL-3.0. Models are not included: Krea 2 is under the Krea 2 Community License (commercial use only below USD 1M annual revenue; acceptable-use and content-filtering obligations). Qwen3-VL-4B: Apache-2.0.<h3 id="thank-you">Thank you</h3>To ComfyUI and its contributors, Krea, the Qwen team, Kostiantyn Hrytsuk (KreaPhoton, CyberKrea), Auryg, the authors of NAG, PAG and SEG, pamparamm, pythongosssss and the whole ComfyUI and Civitai community.Thank you to everyone who will use this, to everyone who will help us, and to the whole community. And if this idea brings even a little progress, we will be happy. ⚗ — AlchemicaMente

⚗ AlchemicaSampler — calibrated Krea 2 Turbo workflow (NAG at cfg 1, two-phase prompts, prompt enhancer)

⚗ First public release. The complete workflow: prompt enhancer (Krea 2's own text encoder) → NAG negative prompt at cfg 1 → calibrated two-stage sampler with two-phase prompts and composition previews → 4 MP output.Needs the custom nodes <a target="_blank" rel="ugc" href="https://github.com/Nessuno76/ComfyUI-AlchemicaSampler">ComfyUI-AlchemicaSampler</a> v1.0.0 (GitHub or ComfyUI-Manager). Other example workflows (simple, pro, calibration, comparison) are in the <code>example_workflows</code> folder of the repository.

Files with SHA256: 54129b2fbfbaba4acabb64d4e248790964f69e1abab188d5fa2c9236832ff77d