Krea2 Dual-Phase Rendering RAW Draft Turbo Finishing Workflow

Watch the full video first if you want to understand how this Krea2 dual-phase rendering workflow works in practice. The video demonstrates a two-stage Krea2 production chain that uses a RAW model to build the initial image and a Turbo model to refine and complete the final result.

This ComfyUI workflow is designed for Krea2 RAW drafting and Turbo finishing. Unlike a simple one-pass Krea2 workflow, this version separates the render into two phases. The first phase focuses on structure, composition, and early visual formation. The second phase uses a Turbo model to tighten the image, improve the final look, and create a more polished output.

The first stage loads krea2_raw_fp8_scaled.safetensors as the RAW base model. This RAW route is passed through Power Lora Loader from rgthree, with multiple LoRA files enabled at strength 0.7. The active LoRA stack includes QJ_Krea2_Lora_E15_mk2, Detailer-KREA2, Neo_Rococo_Krea2_v1, dora, krea2_neondrip, krea2-mikkoph, and krea2_Enhancer. This gives the RAW stage a strong stylized foundation before the image is handed to the second phase.

The second stage loads krea2_turbo_bf16.safetensors as the Turbo finishing model. It also uses a Power Lora Loader stack with the same style-enhancement logic. This allows the final pass to keep the visual direction of the RAW stage while adding a cleaner and more finished Krea2 Turbo surface.

The text encoder route uses qwen3vl_4b_fp8_scaled.safetensors with the Krea2 CLIP type. The VAE route uses qwen_image_vae.safetensors for decoding. The resolution route uses FluxResolutionNode with a 1.5 megapixel target, 9:21 Ultra Tall aspect ratio, and divisible-by-64 alignment. This makes the workflow suitable for tall vertical posters, mobile wallpapers, surreal fashion portraits, fantasy covers, and social-media visual assets.

The Stage 1 sampler uses KSampler with 30 steps, CFG 4, euler sampler, simple scheduler, random seed, and denoise 1. This stage creates the full initial RAW image foundation. The Stage 2 sampler then takes the Stage 1 latent as its input and runs another 30 steps with CFG 1, euler sampler, simple scheduler, fixed seed 666, and denoise 0.42. This second pass is designed to preserve the first-stage composition while refining the final image through the Turbo model.

The workflow also includes an intermediate decode and save path for the first-stage result, so the RAW-stage output can be inspected before the final Turbo refinement. This is useful for comparing how much the second phase changes the image and whether the Turbo finishing pass improves texture, lighting, and final visual impact.

The example prompt focuses on a surreal fashion portrait: a pale ethereal figure in profile, wearing a sculptural black turtleneck coat against a soft gray-green background. Large translucent pink goldfish float around the head, while a biomechanical-organic headdress made from pale hand-like forms covers the upper face. The scene is designed as a luxury conceptual editorial mixed with magical surreal illustration, with elegant negative space, poetic body-horror ornamentation, refined texture, and dreamlike futuristic style.

Main features:

• Krea2 dual-phase rendering workflow

• RAW-stage image foundation

• Turbo-stage final refinement

• krea2_raw_fp8_scaled.safetensors RAW model

• krea2_turbo_bf16.safetensors Turbo model

• Power Lora Loader by rgthree

• Multiple LoRA stack support

• LoRA strength 0.7 style enhancement

• qwen3vl_4b_fp8_scaled.safetensors text encoder

• qwen_image_vae.safetensors VAE

• Stage 1: 30 steps, CFG 4, denoise 1

• Stage 2: 30 steps, CFG 1, denoise 0.42

• euler sampler

• simple scheduler

• RAW-to-Turbo latent handoff

• Intermediate Stage 1 preview and save route

• 1.5 megapixel resolution route

• 9:21 Ultra Tall vertical format

• Final VAEDecode, PreviewImage, and SaveImage output

• Suitable for surreal portraits, fashion editorials, fantasy posters, and Krea2 model-handoff testing

Suggested workflow:

Use this workflow when a single Krea2 render feels either too raw or too locked into one model behavior. Let the RAW stage build the initial image structure, then let the Turbo stage refine the result. Start with a prompt that has strong composition, clear subject identity, controlled color palette, and a defined final style. Because the LoRA stack is already strong, avoid adding too many unrelated style keywords. Use the intermediate preview to judge whether the RAW stage is already strong. If the Turbo stage changes too much, lower the second-stage denoise. If the final result still feels unfinished, strengthen the subject, lighting, and material details in the prompt before changing the model chain.

⚙️ RunningHub Workflow

Try the workflow online right now — no installation required.
👉 Workflow: https://www.runninghub.ai/ai-detail/2073386021416431617?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/2073386021416431617?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。