Fine-tune image generation models

​

Scope

• Supported region: Dokumen ini hanya berlaku untuk wilayah Singapura. Anda harus menggunakan API key dari wilayah ini.
• Supported fine-tuning method: Fine-tuning efisien SFT-LoRA.
• Supported models:
  • Image generation (text-to-image/image-to-image): wan2.7-image-pro, wan2.7-image.

​

How to fine-tune a model

• Text-to-image
• Image-to-image

Fine-tuning objective: Train a character LoRA model.Hasil yang diharapkan: Diberikan prompt teks, model menghasilkan gambar karakter tertentu dalam adegan yang dijelaskan oleh prompt tersebut.

Input promptA person in a crowded morning rush hour subway car, holding onto the handrail, with blurred passengers in the background and tunnel lights visible through the windows, wearing an ordinary office worker white shirt and black trousers, standing facing the camera, half-body shot, realistic candid feel.

Output image (before fine-tuning - text-to-image) Tanpa gambar referensi, model tidak dapat menghasilkan karakter tertentu.

Output image (after fine-tuning) Setelah fine-tuning, model dapat mereproduksi karakter tertentu dari set pelatihan secara konsisten.

Fine-tuning objective: Train a "post-apocalyptic red-black mech armor" LoRA model.Hasil yang diharapkan: Diberikan gambar karakter, model menghasilkan versi bergaya "post-apocalyptic red-black mech armor" dari karakter tersebut tanpa memerlukan prompt teks.

Input image

Output image (before fine-tuning) Prompt teks saja tidak dapat menghasilkan efek "post-apocalyptic red-black mech armor" secara konsisten setiap kali.

Output image (after fine-tuning) Setelah fine-tuning, model dapat mereproduksi efek "post-apocalyptic red-black mech armor" dari set pelatihan tanpa memerlukan prompt teks.

​

Step 1: Upload the dataset

• Image generation - text-to-image: wan-image-t2i-training-dataset.zip
• Image generation - image-to-image: wan-image-i2i-training-dataset.zip

Contoh ini menggunakan text-to-image dan hanya mengunggah set pelatihan. Sistem secara otomatis membagi sebagian dari set pelatihan sebagai set validasi.

curl --location --request POST 'https://dashscope-intl.aliyuncs.com/compatible-mode/v1/files' \
--header "Authorization: Bearer $DASHSCOPE_API_KEY" \
--form 'file=@"./wan-image-t2i-training-dataset.zip"' \
--form 'purpose="fine-tune"'

{
    "id": "file-ft-3c67043a747c-xxxxxx",
    "object": "file",
    "bytes": 73310369,
    "filename": "wan-image-t2i-training-dataset.zip",
    "purpose": "fine-tune",
    "status": "processed",
    "created_at": 1782271893
}

​

Step 2: Fine-tune the model

Step 2.1: Create a fine-tuning job

curl --location 'https://dashscope-intl.aliyuncs.com/api/v1/fine-tunes' \
--header "Authorization: Bearer $DASHSCOPE_API_KEY" \
--header 'Content-Type: application/json' \
--data '{
    "model": "wan2.7-image-pro",
    "training_datasets": [
        {
            "data_source_type": "file_id",
            "file_id": "<replace_with_training_dataset_file_id>"
        }
    ],
    "training_type": "efficient_sft",
    "hyper_parameters": {
        "learning_rate": 3e-5,
        "max_steps": 800,
        "eval_steps": 200,
        "max_token_length": "1k",
        "gradient_clip": 0.5,
        "weight_decay": 0.02,
        "max_pixels": "1k",
        "val_img_size": "1k",
        "generation_type": "t2i",
        "lora_rank": 32,
        "save_total_limit": 10
    }
}'

• job_id: ID pekerjaan, digunakan untuk menanyakan progres.
• finetuned_output: Nama model hasil fine-tuning. Anda harus menggunakan nama ini untuk penerapan dan pemanggilan selanjutnya.
• status: Status pelatihan. Setelah membuat pekerjaan fine-tuning, status awalnya adalah PENDING, yang menunjukkan bahwa pelatihan belum dimulai.

{
    ...
    "output": {
        "job_id": "ft-202511111122-xxxx",
        "status": "PENDING",
        "finetuned_output": "xxxx-ft-202511111122-xxxx",
        ...
    }
}

Step 2.2: Query the fine-tuning job status

curl --location 'https://dashscope-intl.aliyuncs.com/api/v1/fine-tunes/<replace_with_fine_tuning_job_id>' \
--header "Authorization: Bearer $DASHSCOPE_API_KEY" \
--header 'Content-Type: application/json'

• status: Ketika nilainya berubah menjadi SUCCEEDED, pelatihan model selesai dan Anda dapat melanjutkan ke penerapan model.
• usage: Jumlah total token yang dikonsumsi selama pelatihan model, digunakan untuk penagihan.

{
    ...
    "output": {
        "job_id": "ft-202511111122-xxxx",
        "status": "SUCCEEDED",
        "usage": 432000,
        ...
    }
}

​

Step 3: Deploy the fine-tuned model

Step 3.1: Deploy the model as an online service

curl --location 'https://dashscope-intl.aliyuncs.com/api/v1/deployments' \
--header "Authorization: Bearer $DASHSCOPE_API_KEY" \
--header 'Content-Type: application/json' \
--data '{
    "model_name": "<replace_with_model_name>",
    "capacity": 1,
    "plan": "lora"
}'

• deployed_model: Nama model yang diterapkan, digunakan untuk menanyakan status penerapan dan memanggil model.
• status: Status penerapan model. Setelah menerapkan model hasil fine-tuning, status awalnya adalah PENDING, yang menunjukkan bahwa penerapan belum dimulai.

{
    ...
    "output": {
        "deployed_model": "wan2.7-image-pro-xxxxxxxxxxxx",
        "status": "PENDING",
        ...
    }
}

Step 3.2: Query the deployment status

curl --location 'https://dashscope-intl.aliyuncs.com/api/v1/deployments/<replace_with_deployed_model>' \
--header "Authorization: Bearer $DASHSCOPE_API_KEY" \
--header 'Content-Type: application/json'

• status: Ketika status berubah menjadi RUNNING, model telah berhasil diterapkan dan Anda dapat mulai memanggilnya.
• deployed_model: Nama model yang diterapkan.

{
    ...
    "output": {
        "status": "RUNNING",
        "deployed_model": "wan2.7-image-pro-xxxxxxxxxxxx",
        ...
    }
}

​

Step 4: Invoke the model to generate images

curl --location 'https://dashscope-intl.aliyuncs.com/api/v1/services/aigc/image-generation/generation' \
--header 'Content-Type: application/json' \
--header "Authorization: Bearer $DASHSCOPE_API_KEY" \
--header "X-DashScope-Async: enable" \
--data '{
    "model": "<replace_with_deployed_model>",
    "input": {
        "messages": [
            {
                "role": "user",
                "content": [
                    {"text": "s86b5p, A person in a crowded morning rush hour subway car, holding onto the handrail, with blurred passengers in the background and tunnel lights visible through the windows, wearing an ordinary office worker white shirt and black trousers, standing facing the camera, half-body shot, realistic candid feel."}
                ]
            }
        ]
    },
    "parameters": {
        "size": "2K",
        "n": 1
    }
}'

curl --location 'https://dashscope-intl.aliyuncs.com/api/v1/services/aigc/image-generation/generation' \
--header 'Content-Type: application/json' \
--header "Authorization: Bearer $DASHSCOPE_API_KEY" \
--header "X-DashScope-Async: enable" \
--data '{
    "model": "<replace_with_deployed_model>",
    "input": {
        "messages": [
            {
                "role": "user",
                "content": [
                    {"image": "<replace_with_reference_image_URL>"},
                    {"text": "s86b5p, Change the background to an elevator with red lighting. Change the character clothing to red tight-fitting mech armor with black stripe decorations."}
                ]
            }
        ]
    },
    "parameters": {
        "size": "2K",
        "n": 1
    }
}'

{
    "request_id": "4909100c-7b5a-9f92-bfe5-xxxxxx",
    "output": {
        "task_id": "0385dc79-5ff8-4d82-bcb6-xxxxxx",
        "task_status": "PENDING"
    }
}

curl -X GET https://dashscope-intl.aliyuncs.com/api/v1/tasks/86ecf553-d340-4e21-xxxxxxxxx \
--header "Authorization: Bearer $DASHSCOPE_API_KEY"

{
    "request_id": "3f2ebb4e-3d47-97b5-xxxx-xxxxxx",
    "output": {
        "task_id": "aeea547c-e24e-4acb-xxxx-xxxxxx",
        "task_status": "SUCCEEDED",
        "submit_time": "2026-05-29 17:35:23.826",
        "scheduled_time": "2026-05-29 17:35:23.865",
        "end_time": "2026-05-29 17:36:32.498",
        "finished": true,
        "choices": [
            {
                "finish_reason": "stop",
                "message": {
                    "role": "assistant",
                    "content": [
                        {
                            "image": "https://dashscope-7c2c.oss-accelerate.aliyuncs.com/xxx.png?Expires=xxxxxx"
                        }
                    ]
                }
            }
        ]
    },
    "usage": {
        "size": "2048*2048",
        "total_tokens": 770,
        "image_count": 1,
        "output_tokens": 691,
        "input_tokens": 79
    }
}

​

Build custom datasets

​

Dataset format

​

Training set: Required

wan-image-t2i-training-dataset.zip
├── data.jsonl      # Harus bernama data.jsonl, maksimum 20MB
├── 1_0.png         # Gambar target pelatihan, resolusi maks 4096*4096, maks 20MB per gambar, mendukung PNG/JPG/JPEG/WEBP/BMP
├── 1_1.png         # Nama file hanya mendukung karakter Inggris, struktur datar (tanpa subdirektori)
└── 1_2.png

{
  "prompt": "s86b5p, A person in a quiet private library on a peaceful afternoon, with tall dark walnut bookshelves behind them, sunlight streaming through venetian blinds casting striped shadows, wearing a soft beige cable-knit sweater, standing facing the camera, half-body shot, the image has a delicate film grain texture.",
  "img_path": "./1_0.png"
}

wan-image-i2i-training-dataset.zip
├── data.jsonl      # Harus bernama data.jsonl, maksimum 20MB
├── 1_0.jpg         # Gambar target pelatihan (output)
├── 1_1.jpg         # Gambar referensi (input)
├── 6_0.jpg         # Gambar target pelatihan (output)
└── 6_1.jpg         # Gambar referensi (input)

{
  "prompt": "s86b5p, Change the background to an elevator with red lighting, featuring large floor-to-ceiling windows. Change the character's clothing to red tight-fitting mech armor with black stripe decorations.",
  "input_img": "./1_1.jpg",
  "img_path": "./1_0.jpg"
}

wan-image-multi-i2i-training-dataset.zip
├── data.jsonl      # Harus bernama data.jsonl, maksimum 20MB
├── 1_0.jpg         # Gambar target pelatihan (output)
├── 1_ref.jpg       # Gambar referensi 1 (misalnya, foto karakter)
├── 1_pose.jpg      # Gambar referensi 2 (misalnya, gambar pose)
├── 6_0.jpg         # Gambar target pelatihan (output)
├── 6_ref.jpg       # Gambar referensi 1
└── 6_pose.jpg      # Gambar referensi 2

{
  "prompt": "s86b5p, Change the background to an elevator with red lighting, featuring large floor-to-ceiling windows. Outside the windows, there is a post-apocalyptic scene with red mist. Change the character's clothing to red tight-fitting mech armor with black stripe decorations. Standing with both arms stretched horizontally to form a T-shape.",
  "input_imgs": ["./1_ref.jpg", "./1_pose.jpg"],
  "img_path": "./1_0.jpg"
}

​

Validation set: Optional

• Set validasi:
  • Text-to-image: wan-image-t2i-valid-dataset.zip
  • Image-to-image: wan-image-i2i-valid-dataset.zip
• Struktur direktori paket zip:

wan-image-i2i-valid-dataset.zip
├── data.jsonl       # Harus bernama data.jsonl, maksimum 20MB
├── input_001.png    # Opsional, gambar referensi untuk mode image-to-image
└── input_002.png

• File anotasi (data.jsonl): Setiap baris merepresentasikan satu sampel validasi dan harus berupa objek JSON.
  • Text-to-image
  • Image-to-image
  • Multi-image-to-image

  Copy

  {
      "prompt": "s86b5p, A person in a crowded morning rush hour subway car, holding onto the handrail, with blurred passengers in the background and tunnel lights visible through the windows, wearing an ordinary office worker white shirt and black trousers, standing facing the camera, half-body shot, realistic candid feel."
  }
  

  Copy

  {
      "prompt": "s86b5p, Change the background to an elevator with red lighting, featuring large floor-to-ceiling windows. Change the character's clothing to red tight-fitting mech armor with black stripe decorations.",
      "input_img": "./input_001.png"
  }
  

  Set validasi multi-image-to-image menggunakan input_imgs (array) untuk meneruskan beberapa jalur gambar referensi, mendukung hingga 9 gambar.

  Copy

  {
      "prompt": "s86b5p, Change the background to an elevator with red lighting, featuring large floor-to-ceiling windows. Outside the windows, there is a post-apocalyptic scene with red mist. Change the character's clothing to red tight-fitting mech armor with black stripe decorations. Standing with both arms stretched horizontally to form a T-shape.",
      "input_imgs": ["./input_001.png", "./input_002.png"]
  }
  

​

Data scale and limits

• Data volume: Kami merekomendasikan menyediakan minimal 25 gambar (50 atau lebih direkomendasikan untuk hasil yang lebih baik). Gunakan karakter atau gaya yang sama di berbagai adegan dan sudut pandang dengan deskripsi konten yang konsisten.
• Zip package: Saat mengunggah melalui API, ukuran total paket tidak boleh lebih dari 1 GB.
• Training image requirements:
  • Format gambar yang didukung: BMP, JPEG, PNG, dan WEBP.
  • Resolusi gambar tidak boleh lebih dari 4096×4096.
  • Ukuran file gambar individual tidak boleh lebih dari 20 MB.

​

Data collection and cleaning

1. Determine the fine-tuning scenario

• IP character stylization: Latih model untuk mempelajari gaya menggambar karakter IP tertentu, seperti karakter anime atau gambar maskot.
• Fixed visual style: Tingkatkan kemampuan model untuk mereproduksi gaya seni tertentu, seperti ilustrasi datar, lukisan tinta, atau seni piksel.
• Specific scene generation: Replikasi pola komposisi atau templat adegan tertentu, seperti gambar tampilan produk atau tata letak poster.

2. Obtain raw materials

• Generasi AI dan seleksi: Gunakan model dasar Wan untuk menghasilkan gambar secara massal, lalu pilih secara manual sampel berkualitas tinggi yang paling sesuai dengan efek target. Ini adalah metode yang paling umum digunakan.
• Fotografi nyata: Jika tujuan Anda adalah mencapai adegan yang sangat realistis (seperti foto produk nyata atau fotografi potret), menggunakan rekaman hasil jepretan nyata adalah pilihan terbaik.
• Rendering perangkat lunak 3D: Untuk adegan yang memerlukan kontrol detail halus atau gaya rendering 3D, kami merekomendasikan menggunakan perangkat lunak 3D (seperti Blender atau C4D) untuk membuat materi sumber.

3. Clean the data

​

Image annotation: Writing prompts for images

​

Prompt writing formula

​

Evaluate models with validation sets

​

Specify the validation set

Metode 1: Tidak mengunggah set validasi (pembagian otomatis sistem)

Metode 2: Mengunggah set validasi secara manual (ditentukan melalui validation_file_ids)

{
    "model":"wan2.7-image-pro",
    "training_file_ids":[ "<training_set_file_id>" ],
    "validation_file_ids": [ "<custom_validation_set_file_id>" ],
    ...
}

​

Select the best checkpoint for deployment

• Cara mengevaluasi: Nilai hasilnya dengan langsung mengamati gambar pratinjau yang dihasilkan.
• Kriteria seleksi: Temukan checkpoint dengan hasil terbaik dan gaya yang paling sesuai.

​

Procedure

curl --location 'https://dashscope.aliyuncs.com/api/v1/fine-tunes/<replace_with_fine_tuning_job_id>/validation-results' \
--header "Authorization: Bearer $DASHSCOPE_API_KEY" \
--header 'Content-Type: application/json'

{
    "request_id": "da1310f5-5a21-4e29-99d4-xxxxxx",
    "output": [
        {
            "checkpoint": "checkpoint-160"
        },
        ...
    ]
}

curl --location 'https://dashscope-intl.aliyuncs.com/api/v1/fine-tunes/<replace_with_fine_tuning_job_id>/validation-details/<replace_with_checkpoint_to_export>?page_no=1&page_size=10' \
--header "Authorization: Bearer $DASHSCOPE_API_KEY"

{
    "request_id": "375b3ad0-d3fa-451f-b629-xxxxxxx",
    "output": {
        "page_no": 1,
        "page_size": 10,
        "total": 5,
        "list": [
            {
                "img_path": "https://finetune-result.oss-cn-wulanchabu.aliyuncs.com/xxx.png?Expires=xxxxxx",
                "prompt": "s86b5p, Change the background to an elevator equipped with a white ceiling lighting, featuring large floor-to-ceiling windows. Change the character's clothing to red tight-fitting mech armor with black stripe decorations.",
                "input_img": "https://finetune-result.oss-cn-wulanchabu.aliyuncs.com/val_dataset/input_001.png?Expires=xxxxxx"
            },
            ...
        ]
    }
}

curl --location 'https://dashscope-intl.aliyuncs.com/api/v1/fine-tunes/<replace_with_fine_tuning_job_id>/export/<replace_with_checkpoint_to_export>?model_name=<replace_with_exported_model_display_name>' \
--header "Authorization: Bearer $DASHSCOPE_API_KEY"

{
    "request_id": "0817d1ed-b6b6-4383-9650-xxxxx",
    "output": true
}

curl --location 'https://dashscope-intl.aliyuncs.com/api/v1/fine-tunes/<replace_with_fine_tuning_job_id>/checkpoints' \
--header "Authorization: Bearer $DASHSCOPE_API_KEY"

{
    "request_id": "b0e33c6e-404b-4524-87ac-xxxxxx",
    "output": [
         ...,
        {
            "create_time": "2025-11-11T13:27:29",
            "full_name": "ft-202511111122-496e:checkpoint-160",
            "job_id": "ft-202511111122-496e",
            "checkpoint": "checkpoint-160",
            "model_name": "xxxx-ft-202511111122-xxxx-c160", // Bidang penting, digunakan untuk penerapan dan pemanggilan model
            "model_display_name": "xxxx-ft-202511111122-xxxx",
            "status": "SUCCEEDED" // Checkpoint berhasil diekspor
        },
        ...

    ]
}

​

Billing

• Model training: Charged.
  • Biaya = Total token pelatihan × Harga satuan. Lihat Training and deployment pricing.
  • Setelah pelatihan selesai, periksa jumlah total token yang dikonsumsi selama pelatihan dalam bidang usage dari API Retrieve a fine-tuning job.
  Tabel berikut mencantumkan jumlah langkah pelatihan dan perkiraan biaya untuk wan2.7-image, wan2.7-image-pro. Data ini hanya untuk referensi. Hasil pelatihan aktual mengikuti pengiriman akhir, dan biaya mengikuti tagihan resmi. Untuk rumus penagihan detail, lihat Training and deployment pricing.

  generation_type	Image Resolution	Common Step Count	Estimated Token Consumption	Estimated Cost
  t2i (text-to-image)	1K	500	6.400.000	$96
  		1.000	12.800.000	$192
  		2.000	25.600.000	$384
  	2K	500	11.610.000	$174,15
  		1.000	23.220.000	$348,3
  		2.000	46.440.000	$696,6
  i2i (image-to-image)	1K	500	11.610.000	$174,15
  		1.000	23.220.000	$348,3
  		2.000	46.440.000	$696,6
  	2K	500	16.000.000	$240
  		1.000	32.000.000	$480
  		2.000	64.000.000	$960

• Model deployment and invocation: Penerapan gratis. Pemanggilan dikenai biaya sesuai tarif standar model dasar hasil fine-tuning.

  Model ID	LoRA Deployment & Invocation Price
  wan2.7-image-pro	$0,075/image
  wan2.7-image	$0,03/image

​

API reference

​

FAQ

​

Q: How do I design a good trigger word?

• Kami merekomendasikan menggunakan kombinasi karakter langka yang tidak memiliki makna semantik sebenarnya, seperti s86b5p, m01aa, atau EVEAven638123. Pastikan tidak ada makna semantik dalam kosakata model dasar.
• Hindari menggunakan kata bahasa Inggris umum (seperti beautiful, fire, atau dance), karena hal ini akan mencemari pemahaman asli model terhadap kata-kata tersebut.