Watch the full video first if you want to understand how this Krea2 giant image production engine works in practice. The video demonstrates a complete Krea2 image workflow that does not stop at the first render. Instead, it builds a base image first, then sends the result into a dedicated 4K / 6K high-resolution enlargement chain for a cleaner, larger, and more production-ready final output.

This ComfyUI workflow is designed for Krea2 large-format image creation. Its core purpose is to solve a common limitation of fast Krea2 workflows: the first render may look good, but it may not be large enough for posters, covers, print-style artwork, detailed social-media crops, or high-resolution visual presentation. This workflow adds a structured upscale route after the initial Krea2 generation, making it suitable for creators who need a stronger final image instead of only a preview-level output.

The first stage uses Krea2 as the base image engine. The workflow loads krea2_turbo_fp8.safetensors as the main model, qwen3vl_4b_fp8_scaled.safetensors as the Krea2 text encoder, and qwen_image_vae.safetensors for decoding. The base KSampler route uses 8 steps, CFG 1, er_sde sampler, simple scheduler, full denoise, and randomized seed. This first stage is responsible for building the main subject, composition, lighting, color mood, and surreal visual concept.

The example prompt focuses on an ethereal surreal portrait: a calm human figure emerging from clouds, a sculptural face, pale blue porcelain skin, an opened head containing a dream garden, flowers, mist, water, translucent profile silhouettes, luminous goldfish, pastel tones, and poetic high-end illustration texture. This kind of prompt is well suited for a giant upscale workflow because it contains many visual details that benefit from high-resolution refinement.

After the first render is decoded, the workflow enters the high-resolution chain. It includes a CR Upscale Image route using RealESRGAN_x4plus.pth, designed for quick image enlargement and preview-quality enhancement. It also includes a heavier tiled upscale route built around UltimateSDUpscale.

The UltimateSDUpscale section uses 4x_NMKD-Superscale-SP_178000_G.pth as the upscale model. The upscale factor is controlled through a connected value node and is set to 3x. The refinement route uses Krea2 model conditioning again, with euler sampler, simple scheduler, CFG 1, and denoise 0.51. The workflow also uses large tile dimensions, including 4608×6144-related tiled parameters, making it suitable for 4K / 6K-style final image production.

The graph includes MathExpression nodes for dynamic step and tile-related calculations, plus an rgthree Image Comparer node for before-and-after inspection. This makes it practical to compare the lighter upscale route and the heavier tiled upscale route directly.

Main features:

• Krea2 giant image production workflow

• 4K / 6K high-resolution enlargement chain

• Krea2 Turbo FP8 base generation

• krea2_turbo_fp8.safetensors support

• qwen3vl_4b_fp8_scaled.safetensors text encoder

• qwen_image_vae.safetensors VAE

• 8-step base Krea2 generation

• CFG 1 setup

• er_sde sampler

• simple scheduler

• Randomized seed generation

• CR Upscale Image route

• RealESRGAN_x4plus.pth upscale support

• UltimateSDUpscale tiled refinement route

• 4x_NMKD-Superscale-SP_178000_G.pth upscale model

• 3x high-resolution upscale control

• denoise 0.51 tiled refinement

• 4608×6144 large-format tile parameter setup

• MathExpression dynamic parameter control

• rgthree Image Comparer before-and-after comparison

• PreviewImage and SaveImage output

• Suitable for posters, surreal portraits, fantasy art, covers, and large-format AI artwork

Suggested workflow:

Start with a prompt that already produces a strong base image. The first Krea2 render should solve composition, subject, lighting, and visual concept. The upscale chain is for resolution, texture, and final polish, not for fixing a weak image. Use the CR Upscale route for a lighter enhancement preview. Use the UltimateSDUpscale route when the image deserves a heavier 4K / 6K-style finish. Keep denoise around the balanced range when preserving the original render matters. Increase refinement only when the base image lacks texture or needs stronger detail reconstruction. This workflow is best used for images that need to become finished assets rather than quick drafts.

⚙️ RunningHub Workflow

Try the workflow online right now — no installation required.
👉 Workflow: https://www.runninghub.ai/ai-detail/2073337248245960705?inviteCode=rh-v1111

If the results meet your expectations, you can later deploy it locally for customization.

🎁 Fan Benefits: Register to get 1000 points + daily login 100 points — enjoy 4090 performance and 48 GB super power!

📺 Bilibili Updates (Mainland China & Asia-Pacific)

If you’re in the Asia-Pacific region, you can watch the video below to see the workflow demonstration and creative breakdown.
📺 Bilibili Video: https://www.bilibili.com/video/BV1oUTt6ZEEp/

☕ Support Me on Ko-fi

If you find my content helpful and want to support future creations, you can buy me a coffee ☕.
Every bit of support helps me keep creating — just like a spark that can ignite a blazing flame.
👉 Ko-fi: https://ko-fi.com/aiksk

💼 Business Contact

For collaboration or inquiries, please contact aiksk95 on WeChat.

⚙️打开下方链接即可在线体验，无需安装。
👉 工作流： https://www.runninghub.ai/ai-detail/2073337248245960705?inviteCode=rh-v1111
如果觉得效果理想，你也可以在本地进行自定义部署。

🎁 粉丝福利： 注册即送 1000 积分，每日登录 100 积分，畅玩 4090 体验 48 G 超级性能！

📺 Bilibili 更新（中国大陆及南亚太地区）

如果你在中国大陆或南亚太地区，可以通过下方视频查看该工作流的实测效果与构思讲解。
📺 B站视频： https://www.bilibili.com/video/BV1oUTt6ZEEp/

我会在 夸克网盘 持续更新模型资源：
👉 https://pan.quark.cn/s/20c6f6f8d87b
这些资源主要面向本地用户，方便进行创作与学习。