Generate Hyper‑Realistic Journey Images with Flux AI on Alibaba Cloud Serverless

Overview

Flux is a high‑resolution text‑to‑image model from Black Forest Labs. It delivers strong prompt adherence, supports many aspect ratios, and produces realistic, anime, and animal images.

Prerequisites

Enable Function Compute (FC) on Alibaba Cloud.

Enable NAS (file storage) for model and LoRA files.

Free trial quotas

FC free quota: https://common-buy.aliyun.com/package?planCode=package_fcfreecu_cn

NAS free quota: https://free.aliyun.com/?spm=5176.59209.J_5834642020.4.dd5f76b97HqUVE&productCode=nas

Deploying the Flux model

Open the Function Compute console (version 3.0).

If first time, create the default role AliyunFcDefaultRole .

Navigate to Application → Create Application .

Select Artificial Intelligence → Flux Hyper‑Realistic Text‑to‑Image Model and click Create Immediately . Switch to the 3.0 console if the model is not visible.

Choose Direct Deploy and confirm the role has required permissions.

Ensure NAS is enabled; LoRA files will be stored there.

Select a region close to you (East China 1, East China 2, or Japan Tokyo).

Click Create and Deploy Default Environment and wait ~1 minute for deployment and generation of an access domain.

Using ComfyUI with the deployed model

Download the prepared workflow zip from https://labfileapp.oss-cn-hangzhou.aliyuncs.com/FC/v2FLUX%E8%A5%BF%E6%B8%B8%E5%86%8D%E7%8E%B0workflow.zip and extract the JSON file.

In ComfyUI, click Load and import the JSON workflow.

Press Queue Prompt to generate the first image (cold start ~30 s; subsequent images 2‑5 s).

Modify the positive prompt (green box) to generate different characters. Example prompts:

Sun Wukong:

wukong,1 monkey,Solo,Hairy,Covered with hair,Chinese style,armor,looking at viewer,Simple background,cartoon

Pig Bajie:

1 black pig,Solo,Hairy,Covered with hair,Chinese style,armor,looking at viewer,Simple background,Realistic,Portrait

Ox King:

1 black buffalo,Solo,Hairy,Covered with hair,Chinese style,armor,looking at viewer,Simple background,Realistic,Portrait

Optional: experiment with other styles by changing the positive prompt, e.g., “realistic minion holding a sign …” or “a pretty little girl, reaching out to greet …”.

Important operational notes

Protect the generated domain name; sharing it may incur unexpected charges.

The domain ***.devsapp.net is provided for learning and will be reclaimed after 30 days.

ComfyUI keeps a WebSocket connection open, which consumes resources; close the page when not in use.

Conclusion

After completing the steps, you can generate hyper‑realistic images with the Flux model in a serverless environment and explore custom prompts or alternative LoRA models via the provided workflow.