Watch the full video first if you want to understand how this Krea2 three-stage rendering workflow works in practice. The video shows how a Krea2 Turbo image can be built step by step: first creating the base composition, then refining the latent image after the first upscale, and finally polishing the result through a third controlled refinement pass.

This ComfyUI workflow is designed for Krea2 three-stage image rendering. Compared with a simple one-pass Krea2 workflow, this version does not stop after the first generation. It uses a staged latent pipeline to gradually improve the image structure, scale, clarity, and final visual finish. This makes it suitable for creators who want a cleaner and more production-ready result from Krea2 without building a heavy external upscaling system.

The workflow uses krea2_turbo_bf16.safetensors as the main generation model. The text encoder route uses qwen3vl_4b_fp8_scaled.safetensors with the Krea2 CLIP type, and the final decoding route uses qwen_image_vae.safetensors. This keeps the workflow compact, fast, and practical for RunningHub online use.

The prompt route is strengthened through ConditioningKrea2Rebalance. In this workflow, Rebalance is set to the balanced preset with a custom 12-layer weight structure:

1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 2.5, 5.0, 1.1, 4.0, 1.0

The multiplier is set to 2.0, with renormalize enabled. This makes the conditioning stronger than a plain Krea2 render, but less aggressive than the heavy multiplier 4.0 workflows. It is a balanced setup for staged refinement, where the goal is not only strong prompt response, but also cleaner composition and stable detail across multiple passes.

The first stage starts from an EmptyLatentImage controlled by FluxResolutionNode. The active output format is set to 3:4 Golden Ratio, making this workflow suitable for vertical posters, character images, fantasy covers, mobile thumbnails, and stylized AI artwork. Stage 1 uses 6 steps, CFG 1, euler sampler, sgm_uniform scheduler, fixed seed, and full denoise. This stage creates the base image composition.

After Stage 1, the latent is enlarged through LatentUpscaleBy using nearest-exact at 1.25x scale. Stage 2 then runs another 6-step KSampler pass with denoise 0.45. This stage refines the upscaled latent while preserving the main structure.

The workflow repeats the same logic once more. The Stage 2 latent is again upscaled by 1.25x, then Stage 3 performs the final 6-step refinement with denoise 0.45. This gives the image a cleaner final surface, more controlled details, and a more polished high-resolution look.

Main features:

• Krea2 three-stage rendering workflow

• Stage 1 base composition generation

• Stage 2 latent upscale refinement

• Stage 3 final high-resolution polish

• krea2_turbo_bf16.safetensors support

• qwen3vl_4b_fp8_scaled.safetensors text encoder

• qwen_image_vae.safetensors VAE

• ConditioningKrea2Rebalance balanced setup

• Custom 12-layer Rebalance weights

• Multiplier 2.0 with renormalize enabled

• 6 steps per stage

• CFG 1 setup

• euler sampler

• sgm_uniform scheduler

• Fixed seed for repeatable testing

• Two 1.25x latent upscale passes

• nearest-exact latent scaling

• 3:4 Golden Ratio output route

• Empty negative conditioning route

• VAEDecode final image decoding

• PreviewImage and SaveImage output

• Suitable for posters, fantasy art, cinematic characters, and high-detail Krea2 renders

Suggested workflow:

Start with a strong base prompt. The first stage should already solve the subject, composition, lighting, and atmosphere. The second and third stages are mainly for scale, refinement, and polish, not for completely changing the image. If the final image drifts too far from the original idea, reduce the refinement denoise or simplify the prompt. If the result still looks soft, strengthen texture, lighting, and focal-detail words in the prompt. This workflow works best when the base image is already good and you want a cleaner, more finished Krea2 render through staged latent refinement.

⚙️ RunningHub Workflow

Try the workflow online right now — no installation required.
👉 Workflow: https://www.runninghub.ai/post/2072217224156377090?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/BV1FWT76KENn/

☕ 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/post/2072217224156377090?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/BV1FWT76KENn/

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。