← Docs Home · Model Catalog · HTTP API · Troubleshooting

HiDream-E1-Full LoRA Image Editing with Diffusers (AI Toolkit-trained)

API model id: hidream_e1
URL slug: hidream-e1

This page documents the reference Diffusers editing pipeline for hidream_e1 (HiDream-E1-Full). It is designed for running LoRAs trained with ostris/ai-toolkit while minimizing training preview vs inference mismatch.

Run in the cloud (optional): If you want to reproduce the examples on this page in a pinned runtime without local CUDA/driver setup (and reduce preview‑vs‑inference drift), run it via RunComfy’s Cloud AI Toolkit (Train + Inference). 👉 You can open it here: Cloud AI Toolkit (Train + Inference)

Quick facts

Field	Value
Pipeline	src/pipelines/hidream.py (class HiDreamE1Pipeline)
Base checkpoint	HiDream-ai/HiDream-E1-Full
Defaults	sample_steps=28, guidance_scale=5.0, seed=42
Resolution snapping	Floors width/height to a multiple of 16
Control image	Required (ctrl_img)
LoRA scale behavior	Fused into transformer weights at load time; changing loras[].network_multiplier requires reload.
Needs AI Toolkit	Required (needs a local ostris/ai-toolkit checkout via AI_TOOLKIT_PATH)

Reference implementation (source of truth)

• Pipeline implementation: src/pipelines/hidream.py
• Shared behaviors (snapping, seeding, LoRA base logic): src/pipelines/base.py
• Request schema (parameter names): src/schemas/request.py
• API routes and validation: src/api/v1/inference.py

Minimal API request

{
  "model": "hidream_e1",
  "trigger_word": "sks",
  "prompts": [
    {
      "prompt": "[trigger] make it look like a watercolor illustration",
      "neg": "",
      "seed": 42,
      "width": 1024,
      "height": 1024,
      "sample_steps": 28,
      "guidance_scale": 5.0,
      "ctrl_img": "<base64_or_url>"
    }
  ],
  "loras": [
    {
      "path": "my_lora_job/my_lora.safetensors",
      "network_multiplier": 1.0
    }
  ]
}

Pipeline behavior that matters

• hidream_e1 is an editing pipeline: ctrl_img is mandatory and is resized to the requested output size.
• The LoRA is fused into the transformer at load time (no dynamic adapter scaling).
• The pipeline uses an internal image_guidance_scale (currently fixed in code) in addition to guidance_scale.

Preview-matching notes (training preview vs inference mismatch)

• HiDream is sensitive to its conditioning stack (tokenizers/models/max sequence length). If you use a different stack than the reference pipeline, outputs can drift.
• Width/height are floored to a multiple of 16.
• Because LoRA is fused, changing loras[].network_multiplier requires a pipeline reload.

Related

• HiDream I1 (text-to-image)
• See the Model Catalog for all supported models.