Draw a few flat colour blocks, describe the finished picture, get a finished Krea 2 illustration that keeps your layout. The graph repaints your sketch at three strengths at once, 0.55, 0.62 and 0.70 with the same seed, and gives you a side-by-side sheet so you can pick. Core ComfyUI nodes only, no custom nodes.
It reads colour blocks as objects. A red trapezoid with a skin-coloured circle on top becomes a woman in a red coat; a blue rectangle on a wall becomes a window. Tell the prompt what each block is.
REQUIRED MODELS
<ul>
<li>Diffusion model: <a target="_blank" rel="ugc" href="https://civarchive.com/models/958009?modelVersionId=3139241">RedCraft 3.0 Krea 2</a>, the INT8 convrot build, <code>redcraft23_30Krea2_int8_convrot.safetensors</code>, in <code>models/diffusion_models</code>. Plain Krea 2 and Krea 2 Turbo load too; RedCraft is what I tested.</li>
<li>Text encoder: the Qwen3-VL 4B text encoder from the Krea 2 downloads (see <a target="_blank" rel="ugc" href="https://comfyui-wiki.com/en/tutorial/advanced/image/krea/krea-2">the Krea 2 guide on comfyui-wiki.com</a>); I run a community int8 ConvRot build of it (<code>qwen3vl_4b_int8_convrot</code>), and the bf16 or fp8 file works the same. It goes in <code>models/text_encoders</code>, loaded with CLIPLoader type <code>krea2</code>.</li>
<li>VAE: the Qwen-Image VAE, <code>qwen_image_vae.safetensors</code>, in <code>models/vae</code>, from <a target="_blank" rel="ugc" href="https://huggingface.co/Comfy-Org/Qwen-Image_ComfyUI/tree/main/split_files/vae">Comfy-Org on Hugging Face</a>.</li>
</ul>
HOW TO USE IT
<ol>
<li>Load your sketch or blocking in the blue node. It is resized to about 1 megapixel, in steps of 16 px.</li>
<li>In the green node, describe the finished picture in one full paragraph: the medium, what each colour block is, the light. A block you do not name, the model guesses.</li>
<li>Run. You get three files and one sheet: input, 0.55, 0.62, 0.70.</li>
<li>Pick a strength. Change the seed for other versions of the same layout.</li>
</ol>
WHICH STRENGTH
<ul>
<li>0.55 stays closest to your shapes and colours. The style changes and the detail is invented inside your blocks.</li>
<li>0.62 is the middle step. Hair, folds and furniture start to shift.</li>
<li>0.70 redraws the most. The layout held in all three of my tests, but details moved: hair length, the window frame, which way the bed faces.</li>
<li>I did not test below 0.55 or above 0.70. The three denoise values sit on the three KSamplers if you want a different ladder.</li>
</ul>
WHAT I TESTED
Three blockings made in a few minutes with flat colour only, no outlines and no texture: a standing figure, a bedroom in perspective and a mountain lake. Each ran through the graph at the three strengths. I compared the 32x32 brightness layout of the input with each result (1.0 means identical): figure 0.99 / 0.97 / 0.95, room 0.98 / 0.96 / 0.91, landscape 0.98 / 0.96 / 0.93 at 0.55 / 0.62 / 0.70. Edge detail grew 2 to 5 times against the flat input, so the style was repainted and not just tinted. Judge the sheets in the gallery with your own eye; the numbers only say the layout stayed.
SETTINGS
<ul>
<li>Steps <code>8</code>. It is a distilled model, more steps do not help. CFG <code>1.0</code>, sampler <code>er_sde</code>, scheduler <code>simple</code>. Decode is <code>VAEDecodeTiled</code>, tile 512, overlap 64. Negative: a <code>ConditioningZeroOut</code>, because a negative prompt does nothing at CFG 1.0.</li>
</ul>
LIMITS
<ul>
<li>Blockings drawn with a big head and a short body give a figure with child proportions. Draw adult proportions, and say "the whole figure about seven and a half heads tall" in the prompt. My first figure came out childlike for that reason.</li>
<li>Tested on three simple inputs and one seed each, not on hand sketches or photographs.</li>
<li>Tested on RedCraft 3.0 (Krea 2) on ROCm (a 7900 XTX). Core nodes only, so CUDA builds should run it as is. The UI file was built by script and checked against the node definitions, not saved from a browser; if a node shows red on load, tell me which. It needs a ComfyUI build where <code>ImageScaleToTotalPixels</code> has the <code>resolution_steps</code> input.</li>
</ul>
If this worked for you, a thumbs-up here helps the next person find it.

Krea 2 Sketch to Render: Flat Colour Blocking to Finished Illustration

Draw a few flat colour blocks, describe the finished picture, get a finished Krea 2 illustration that keeps your layout. The graph repaints your sketch at three strengths at once, 0.55, 0.62 and 0.70 with the same seed, and gives you a side-by-side sheet so you can pick. Core ComfyUI nodes only, no custom nodes.
It reads colour blocks as objects. A red trapezoid with a skin-coloured circle on top becomes a woman in a red coat; a blue rectangle on a wall becomes a window. Tell the prompt what each block is.
REQUIRED MODELS
<ul>
<li>Diffusion model: <a target="_blank" rel="ugc" href="https://civitai.red/models/958009?modelVersionId=3139241">RedCraft 3.0 Krea 2</a>, the INT8 convrot build, <code>redcraft23_30Krea2_int8_convrot.safetensors</code>, in <code>models/diffusion_models</code>. Plain Krea 2 and Krea 2 Turbo load too; RedCraft is what I tested.</li>
<li>Text encoder: the Qwen3-VL 4B text encoder from the Krea 2 downloads (see <a target="_blank" rel="ugc" href="https://comfyui-wiki.com/en/tutorial/advanced/image/krea/krea-2">the Krea 2 guide on comfyui-wiki.com</a>); I run a community int8 ConvRot build of it (<code>qwen3vl_4b_int8_convrot</code>), and the bf16 or fp8 file works the same. It goes in <code>models/text_encoders</code>, loaded with CLIPLoader type <code>krea2</code>.</li>
<li>VAE: the Qwen-Image VAE, <code>qwen_image_vae.safetensors</code>, in <code>models/vae</code>, from <a target="_blank" rel="ugc" href="https://huggingface.co/Comfy-Org/Qwen-Image_ComfyUI/tree/main/split_files/vae">Comfy-Org on Hugging Face</a>.</li>
</ul>
HOW TO USE IT
<ol>
<li>Load your sketch or blocking in the blue node. It is resized to about 1 megapixel, in steps of 16 px.</li>
<li>In the green node, describe the finished picture in one full paragraph: the medium, what each colour block is, the light. A block you do not name, the model guesses.</li>
<li>Run. You get three files and one sheet: input, 0.55, 0.62, 0.70.</li>
<li>Pick a strength. Change the seed for other versions of the same layout.</li>
</ol>
WHICH STRENGTH
<ul>
<li>0.55 stays closest to your shapes and colours. The style changes and the detail is invented inside your blocks.</li>
<li>0.62 is the middle step. Hair, folds and furniture start to shift.</li>
<li>0.70 redraws the most. The layout held in all three of my tests, but details moved: hair length, the window frame, which way the bed faces.</li>
<li>I did not test below 0.55 or above 0.70. The three denoise values sit on the three KSamplers if you want a different ladder.</li>
</ul>
WHAT I TESTED
Three blockings made in a few minutes with flat colour only, no outlines and no texture: a standing figure, a bedroom in perspective and a mountain lake. Each ran through the graph at the three strengths. I compared the 32x32 brightness layout of the input with each result (1.0 means identical): figure 0.99 / 0.97 / 0.95, room 0.98 / 0.96 / 0.91, landscape 0.98 / 0.96 / 0.93 at 0.55 / 0.62 / 0.70. Edge detail grew 2 to 5 times against the flat input, so the style was repainted and not just tinted. Judge the sheets in the gallery with your own eye; the numbers only say the layout stayed.
SETTINGS
<ul>
<li>Steps <code>8</code>. It is a distilled model, more steps do not help. CFG <code>1.0</code>, sampler <code>er_sde</code>, scheduler <code>simple</code>. Decode is <code>VAEDecodeTiled</code>, tile 512, overlap 64. Negative: a <code>ConditioningZeroOut</code>, because a negative prompt does nothing at CFG 1.0.</li>
</ul>
LIMITS
<ul>
<li>Blockings drawn with a big head and a short body give a figure with child proportions. Draw adult proportions, and say "the whole figure about seven and a half heads tall" in the prompt. My first figure came out childlike for that reason.</li>
<li>Tested on three simple inputs and one seed each, not on hand sketches or photographs.</li>
<li>Tested on RedCraft 3.0 (Krea 2) on ROCm (a 7900 XTX). Core nodes only, so CUDA builds should run it as is. The UI file was built by script and checked against the node definitions, not saved from a browser; if a node shows red on load, tell me which. It needs a ComfyUI build where <code>ImageScaleToTotalPixels</code> has the <code>resolution_steps</code> input.</li>
</ul>
If this worked for you, a thumbs-up here helps the next person find it.

Files with SHA256: 7e3605f5a5616e80e6b623bb96f02fab169a2cf2af9aa3bacf351ed533a7f218