Updated: Jun 6, 2026
character
Watch the full video first if you want to understand how this Bernini-R image editing workflow works in practice. The video shows how one source image can be edited through a text instruction, how the workflow expands a simple idea into a stronger Bernini prompt, and how the final result can be exported as a clean single-frame image.
This ComfyUI workflow is designed for Bernini-R image-to-image editing. Its main purpose is to take an existing image, preserve the important visual identity of the original subject, and apply a controlled transformation through text. Compared with pure text-to-image generation, this workflow starts from a real source image, so it can maintain the subject’s face, clothing, composition, visual direction, and key scene structure while changing the pose, background, object interaction, lighting, or atmosphere.
The workflow is built around the Bernini-R high-noise and low-noise dual-model route. It uses Bernini_HIGH_fp8_e4m3fn_scaled.safetensors and Bernini_LOW_fp8_e4m3fn_scaled.safetensors as the two model branches. It also uses UMT5 XXL fp8 text encoding, Wan 2.1 VAE, BerniniConditioning, KSamplerAdvanced, VAEDecode, SaveImage, and PathchSageAttentionKJ. The generation chain also includes LightX2V LoRA and UnifiedReward-Flex LoRA for both high-noise and low-noise stages, helping the workflow improve speed, image quality, and final visual coherence.
The source image section is the foundation of the workflow. LoadImage imports the original image, then image_scale_pixel_v2 prepares the image size and alignment before it enters BerniniConditioning. This makes the workflow suitable for controlled editing tasks such as changing a character’s pose, replacing a background, adding an object, changing the environment, converting the scene style, or creating a more cinematic version of an existing image.
The prompt creation section is also important. BerniniPromptEnhancer is set to the i2i task type, meaning the workflow is optimized for image editing rather than pure generation. The user can write a short edit instruction, and the prompt enhancer builds a Bernini-specific system prompt. RHLLMChatNode then rewrites the task into a more detailed editing prompt. The output is cleaned through StringReplace nodes, removing the JSON wrapper before the final prompt is sent into CLIPTextEncode.
In the uploaded example, the edit instruction changes the scene into a bustling urban city street, adjusts the man’s pose so he is walking forward, and adds the action of holding and eating vanilla ice cream in a waffle cone. At the same time, the prompt preserves the man’s facial features, messy black hair, and traditional blue-grey layered clothing with intricate patterns. This demonstrates the workflow’s main value: changing the scene and action while keeping the original subject recognizable.
The generation route uses BerniniConditioning in i2i mode with a 480×848 image setup and one-frame output logic. The first KSamplerAdvanced stage handles the high-noise transformation, where the main edit is introduced. The second stage performs low-noise refinement, improving detail, stability, and final image polish. The final latent is decoded with Wan 2.1 VAE and exported through SaveImage.
Compared with ordinary image editing workflows, this Bernini-R setup is more structured. It combines source image anchoring, LLM prompt expansion, Bernini task conditioning, dual-stage sampling, SageAttention optimization, acceleration LoRA, reward-aligned LoRA, and clean single-image output into one reusable creator pipeline.
Main features:
Bernini-R image-to-image editing workflow
One source image + text edit instruction
Preserves subject identity and core composition
Supports background, pose, action, and object edits
Bernini HIGH / LOW fp8 dual-model route
UMT5 XXL fp8 text encoder
Wan 2.1 VAE decoding
LoadImage source image input
image_scale_pixel_v2 image preparation
BerniniPromptEnhancer I2I prompt creation
RHLLMChatNode automatic prompt rewriting
JSON cleanup chain for LLM output
BerniniConditioning I2I control
PathchSageAttentionKJ optimization
LightX2V high / low noise LoRA support
UnifiedReward-Flex high / low noise LoRA support
KSamplerAdvanced two-stage generation
SaveImage single-frame output
Suggested workflow:
Prepare one clean source image first. The subject should be readable, the face or main object should not be blocked, and the image should already contain the identity or composition you want to preserve. Load the image into the workflow, then write a direct edit instruction. Describe what should change and what must stay unchanged. For example, define the new background, pose, object, lighting, action, camera angle, and style, while also locking the face, clothing, hairstyle, body shape, or original composition. Let BerniniPromptEnhancer and RHLLMChatNode expand the task into a more complete Bernini editing prompt. Check the cleaned prompt before rendering. If the subject changes too much, strengthen preservation rules. If the edit is too weak, make the target transformation more explicit.
⚙️ RunningHub Workflow
Try the workflow online right now — no installation required.
👉 Workflow: https://www.runninghub.ai/post/2062541633941495810?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/BV1yLEc6dEJc/
☕ 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/2062541633941495810?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/BV1yLEc6dEJc/
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。