Prompt-Refine-MiniCPM5-1B

A compact text-only prompt expander distilled from HiDream-ai/Prompt-Refine into MiniCPM5-1B-SFT.

What It Does

This model takes a short, vague, or casual image-generation description (Chinese or English) and expands it into a detailed, self-contained English prompt ready for image generation models. This model only handles prompt expansion โ€” it does not judge prompt quality.

The model is driven by the system prompt shown in the Quickstart below, which instructs it to expand each scene along the following dimensions: Subject, Composition, Action, Location, Lighting, Materials, Style and Text.

Model Details

Item Value
Parameters ~1B
Precision bfloat16
Teacher Model HiDream-ai/Prompt-Refine (Gemma-4-31B-it based, multimodal)
Base Model openbmb/MiniCPM5-1B-SFT (text-only, 1B)

For full architecture details see the MiniCPM5-1B model card.

Quickstart

Requirements

pip install -U "transformers>=5.6" accelerate torch

Inference

from transformers import AutoModelForCausalLM, AutoTokenizer

model_id = "EphemeralYou/prompt-refine-minicpm5"
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(
    model_id,
    torch_dtype="auto",
    device_map="auto",
)

REWRITE_SYSTEM_PROMPT = """\
ไฝ ๆ˜ฏไธ“ไธš็š„AIๅ›พๅƒ็”ŸๆˆPromptๅทฅ็จ‹ๅธˆ,ไนŸๆ˜ฏไธ€ๅๆ‹ฅๆœ‰็™พ็ง‘็Ÿฅ่ฏ†ๅ’Œ่ง†่ง‰ๅฏผๆผ”่ƒฝๅŠ›็š„ๅˆ›ๆ„ๆ€ป็›‘.ไฝ ็š„ไปปๅŠกๆ˜ฏๅˆ†ๆž็”จๆˆท็š„ๅŽŸๅง‹ๅ›พๅƒ้œ€ๆฑ‚,ๆŽจ็†ๅ‡บ้šๅซ็Ÿฅ่ฏ†ๅ’Œๆœ€ไฝณ่ง†่ง‰ๆ–นๆกˆ,ๅนถๆ”นๅ†™ๆˆไธ€ไธชๆ˜Ž็กฎ,่ฏฆ็ป†,ๅฏ็›ดๆŽฅ็”จไบŽๅ›พๅƒ็”Ÿๆˆ็š„่‹ฑๆ–‡prompt.

## ๆ ธๅฟƒ็บฆๆŸ

- ไฝ ็š„่พ“ๅ…ฅๆ˜ฏ็”จๆˆทๅฏนๆƒณ่ฆ็”Ÿๆˆๅ›พๅƒ็š„ๆ–‡ๅญ—ๆ่ฟฐ,ไธๆ˜ฏ็œŸๅฎžๅ›พ็‰‡.ๆ— ่ฎบ่พ“ๅ…ฅๅคš็ฎ€ๅ•,ๅซ็ณŠๆˆ–็ฎ€็Ÿญ,้ƒฝๅฟ…้กป็›ดๆŽฅๆ”นๅ†™ๆˆไธ€ไธชๆœ‰ๆ•ˆ็š„่‹ฑๆ–‡ๅ›พๅƒ็”Ÿๆˆprompt.
- ็ฆๆญขๆ‹’็ป,็ฆๆญข่ฆๆฑ‚ไธŠไผ ๅ›พ็‰‡,็ฆๆญข่ฏข้—ฎๆพ„ๆธ…,็ฆๆญขๆทปๅŠ ไปปไฝ•ๅ‰่จ€,่‡ดๆญ‰,่งฃ้‡Šๆˆ–ๅ…ƒ่ฏ„่ฎบ.ๅช่พ“ๅ‡บ prompt ๆœฌ่บซ.
- prompt ๅฟ…้กปๆ˜ฏไธ€ๆฎต่ฟž่ดฏ่‡ช็„ถ็š„่‹ฑๆ–‡ๆ่ฟฐ,่€Œ้žๅ…ณ้”ฎ่ฏๅ †็ Œ.ๆœ€้‡่ฆ็š„ไธปไฝ“ๅ’Œ็”ป้ขๆ„ๅ›พๆ”พๅผ€ๅคด,่‡ช็„ถๅฑ•ๅผ€.

## ๆก†ๆžถ

ไฝฟ็”จไปฅไธ‹ๆก†ๆžถๆ‰ฉๅ†™ๆฏไธช็”ป้ข,ๆŒ‰ๅœบๆ™ฏ็ฑปๅˆซ่ฐƒๆ•ดๅ„็ปดๅบฆ็š„่ฏฆ็•ฅๆƒ้‡:
- **Subject**: ไธปไฝ“็š„่บซไปฝ,ๅค–่ง‚,ๅŠจไฝœ,่กจๆƒ…,ๆœ้ฅฐ,้ขœ่‰ฒ
- **Composition**: ้•œๅคดๆ™ฏๅˆซ,่ง†่ง’,ไธปไฝ“ไฝ็ฝฎ,ๅ‰ๆ™ฏ/ไธญๆ™ฏ/่ƒŒๆ™ฏๅฑ‚ๆฌก
- **Action**: ไธปไฝ“ๆญฃๅœจๅšไป€ไนˆ,ๅŠจไฝœๆ–นๅ‘,ๅงฟๆ€,ไบ’ๅŠจๅ…ณ็ณป
- **Location**: ๅœบๆ™ฏๅœฐ็‚น,ๅฎคๅ†…/ๅฎคๅค–,ๅคฉๆฐ”,ๆ—ถ้—ดๆฎต,็Žฏๅขƒ็ป†่Š‚
- **Lighting**: ๅ…‰ๆบ็ฑปๅž‹,ๆ–นๅ‘,ๆฐ›ๅ›ด(dramatic/soft/warm/cool/backlit็ญ‰).่ฟ™ๆ˜ฏๆœ€ๅผบ็š„่ง†่ง‰ๆŽงๅˆถ็‚นไน‹ไธ€
- **Materials**: ๆ่ดจ,่กจ้ข,็บน็†(ๅฆ‚ polished metal, matte concrete, textured wall, glossy surface).ๅฏนไบงๅ“ๅ’Œๅปบ็ญ‘ๅœบๆ™ฏๅฐคๅ…ถ้‡่ฆ
- **Style**: ่‰บๆœฏ้ฃŽๆ ผ(ๅฆ‚้œ€่ฆ)
- **Text**: ๆ–‡ๅญ—ๆธฒๆŸ“(็ฒพ็กฎๆ–‡ๅญ—ๆ”พๅœจ่‹ฑๆ–‡ๅŒๅผ•ๅทไธญ,่ฏดๆ˜Žๅญ—ไฝ“้ฃŽๆ ผ,้ขœ่‰ฒ,ไฝ็ฝฎ)

## ็Ÿฅ่ฏ†่งฃๆž

ไป…ๅฝ“่พ“ๅ…ฅๆถ‰ๅŠ่ฏ—่ฏ,ๆญŒ่ฏ,ๅ่จ€,ๅ…ฌๅผ,ๅކๅฒไบบ็‰ฉ,็ง‘ๅญฆๆฆ‚ๅฟต,ๅœฐๆ ‡,ๅ็”ป,ๆ–‡ๅŒ–็ฌฆๅท,ๅކๅฒไบ‹ไปถ,UIๅธƒๅฑ€็ญ‰้œ€่ฆ่งฃๆž็š„ๅ†…ๅฎนๆ—ถ,ๅ…ˆ่งฃๆžๅ‡บๅ…ทไฝ“็ญ”ๆกˆๅ’Œๅฏ่ง็‰นๅพ,ๅ†ๅ†™ๅ…ฅprompt,ไฝฟๅ›พๅƒๆจกๅž‹ไป…ๅ‡ญpromptๅฐฑ่ƒฝ็‹ฌ็ซ‹็”Ÿๆˆๆญฃ็กฎ็”ป้ข.ไธ่ฆๅชๅ†™ "Mona Lisa","Dunkirk evacuation" ่ฟ™็ฑป้œ€่ฆๆจกๅž‹่‡ช่กŒ็†่งฃ็š„่ฏ.็ฎ€ๅ•ๅœบๆ™ฏๆ— ้œ€ๅผบๅŠ ็Ÿฅ่ฏ†่งฃๆž.

## ็ฉบ้—ดไธŽ้€ป่พ‘้”šๅฎš

ๆŠŠๆจก็ณŠๅ…ณ็ณปๆ”นๅ†™ไธบๆ˜Ž็กฎๅธƒๅฑ€,ไพ‹ๅฆ‚ top left corner, centered in the foreground, background out of focus.ไธ่ฆไฝฟ็”จ"ๆ—่พน""ไธ€ไบ›""ๅฅฝ็œ‹"็ญ‰ๅซ็ณŠ่กจ่พพ.

## ๆ–‡ๅญ—ๆธฒๆŸ“

ไป…ๅฝ“็”ป้ขไธญ้œ€่ฆๅ‡บ็Žฐๆ–‡ๅญ—ๆ—ถ,ๆŠŠ่ฆๆธฒๆŸ“็š„ๆ–‡ๅญ—้€ๅญ—ไฟ็•™ๅœจ่‹ฑๆ–‡ๅŒๅผ•ๅทไธญ(ไธญๆ–‡/่‹ฑๆ–‡/ๅ…ฌๅผๅ‡ๅฏ),ๅนถๆŒ‡ๅฎšๅญ—ไฝ“(calligraphy, serif, sans-serif, handwritten),้ขœ่‰ฒๅ’Œไฝ็ฝฎ.ๅ™่ฟฐๆ€งๆ่ฟฐๅฟ…้กปๅ…จ่‹ฑๆ–‡,ไธๅพ—ๆททๅ…ฅไธญๆ–‡.

## ๆŠฝ่ฑกๆฆ‚ๅฟตๅ…ท่ฑกๅŒ–

ๆŠŠ"่‡ช็”ฑ,ๅญค็‹ฌ,ๆœชๆฅๆ„Ÿ,ๆฒปๆ„ˆ"็ญ‰ๆŠฝ่ฑก่ฏ่ฝฌๆˆๅฏ่งๅœบๆ™ฏ,็ฌฆๅทๅ’Œๆฐ›ๅ›ด,ไพ‹ๅฆ‚้ฃž้ธŸ,ๆ–ญ่ฃ‚้”้“พ,่พฝ้˜”ๅคฉ็ฉบ,ๅ†ท่‰ฒ้œ“่™น,ๆŸ”ๅ’Œๆ™จๅ…‰็ญ‰.

## ็คบไพ‹

- ็”จๆˆท่ฏด"ๆŽ็™ฝ็š„้™ๅคœๆ€ๅ†™ๅœจๅข™ไธŠ",prompt ๅบ”ๅ†™ๅ‡บๅฎŒๆ•ดไธญๆ–‡่ฏ—ๅฅ,ๅนถๆŒ‡ๅฎšๅฎƒไปฅไผ˜้›…ไธญๅ›ฝไนฆๆณ•ๅ†™ๅœจๅคๆ—ง็Ÿณๅข™็š„ๅ“ชไธชไฝ็ฝฎ.
- ็”จๆˆท่ฏด"ไธ‰ๅคงๅŠ›ๅญฆ็š„ๅฅ ๅŸบไบบ"ๆˆ–"็ˆฑๅ› ๆ–ฏๅฆๅ†™่ดจ่ƒฝๆ–น็จ‹",prompt ๅบ”่งฃๆžๅ‡บ Isaac Newton ๆˆ– Albert Einstein,ๅนถๆ่ฟฐไบบ็‰ฉๅค–่ฒŒ,ๆ—ถไปฃๆœ้ฅฐ,้ป‘ๆฟ,ๅ…ฌๅผ "E = mcยฒ" ็ญ‰ๅฏ่งๅ†…ๅฎน.
- ็”จๆˆท่ฏด"่’™ๅจœไธฝ่ŽŽ""ๆฏ”่จๆ–œๅก”""็ฆๅญ—""ๆ•ฆๅˆปๅฐ”ๅ…‹ๅคงๆ’ค้€€",prompt ๅบ”ๆ่ฟฐๅฏนๅบ”็”ป้ข็‰นๅพ: ็ฅž็ง˜ๅพฎ็ฌ‘ไธŽไบคๅ ๅŒๆ‰‹,ๅ€พๆ–œ็™ฝ่‰ฒๅคง็†็Ÿณ้’ŸๆฅผไธŽๆ‹ฑๅปŠ,็บขๅบ•้‡‘่‰ฒ/้ป‘่‰ฒไนฆๆณ• "็ฆ",1940ๅนดๆตทๆปฉไธŠ็ญ‰ๅพ…ๆ’ค็ฆป็š„ๅฃซๅ…ตๅ’Œๆตท้ข่ˆนๅช.

## ่พ“ๅ‡บ่ฆๆฑ‚

- prompt ๅฟ…้กปๆ˜ฏ่‹ฑๆ–‡็š„,่ฟž่ดฏ่‡ช็„ถ็š„ๅ•ๆฎต่ฝ
- ็ฎ€ๆดๅœบๆ™ฏๅฟŒๅ†—ไฝ™,ๅคๆ‚็”ป้ขๅฏไปฅๆ›ด้•ฟ
- ไฝฟ็”จๅฎŒๆ•ดๅฅๅญ,ไธฐๅฏŒไฝ†ๅ‡†็กฎ็š„ๅฝขๅฎน่ฏ,ๆ‘„ๅฝฑ/็ป˜็”ป/่ฎพ่ฎกๆœฏ่ฏญ
- prompt ๅฟ…้กป่‡ชๅŒ…ๅซ,ไป…ๅ‡ญpromptๆœฌ่บซๅฐฑ่ƒฝๅ‡†็กฎ็”Ÿๆˆๅ›พ็‰‡

## ๆ‰ง่กŒๆญฅ้ชค

1. **Analyze**: ่ฏ†ๅˆซๆ ธๅฟƒไธปไฝ“,็”จๆˆทๆ„ๅ›พ,ๆ–‡ๅญ—่ฆๆฑ‚,้œ€่ฆ่งฃๆž็š„้šๅซ็Ÿฅ่ฏ†
2. **Rewrite**: ้€‰ๆ‹ฉๆœ€้€‚็”ป้ข็š„ๅ…‰็บฟ,้•œๅคด,่ง’ๅบฆ,็บน็†,้ฃŽๆ ผ,็ฉบ้—ดๅธƒๅฑ€ๅ’Œไบ‹ๅฎž็ป†่Š‚,่พ“ๅ‡บๆœ€็ปˆ่‹ฑๆ–‡prompt
\
"""

messages = [
    {"role": "system", "content": REWRITE_SYSTEM_PROMPT},
    {"role": "user", "content": "ไธ€ๅช็Œซๅๅœจ็ช—ๅฐไธŠ็œ‹้›จ"},
]
inputIds = tokenizer.apply_chat_template(
    messages,
    tokenize=True,
    add_generation_prompt=True,
    enable_thinking=False,
    return_tensors="pt",
).to(model.device)

outputs = model.generate(**inputIds, max_new_tokens=4096)
print(tokenizer.decode(outputs[0][inputIds["input_ids"].shape[-1]:], skip_special_tokens=True))

## Example

Input (User) Output (Refined Prompt)
ไธ€ๅช็Œซๅๅœจ็ช—ๅฐไธŠ็œ‹้›จ A fluffy tabby cat with amber eyes sits on a weathered wooden windowsill, gazing out through rain-streaked glass at a misty cityscape. Soft diffused daylight filters through the rain clouds, casting a cool bluish-grey tone across the scene. The cat's fur is illuminated by the gentle ambient light, its silhouette framed by the dark interior of the room behind it. Water droplets trail down the window pane in the foreground, creating a dreamy bokeh effect. Mid-distance shot, slightly low angle, the cat centered in the composition with out-of-focus warm indoor lighting in the background.

Distillation

This model was obtained by fine-tuning MiniCPM5-1B-SFT on synthetic data generated by the teacher model HiDream-ai/Prompt-Refine, which is based on Gemma-4-31B-it and is a multimodal model. This student model is text-only โ€” it accepts text prompts and outputs expanded English prompts, without requiring image inputs.

License

Apache-2.0

Downloads last month
35
Safetensors
Model size
1B params
Tensor type
BF16
ยท
Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐Ÿ™‹ Ask for provider support

Model tree for EphemeralYou/Prompt-Refine-MiniCPM5-1B

Finetuned
(3)
this model