Forge API and Flux

[osi1880vr]

Hi There,

I'm kind of a heavy API user of Forge since looong time. I did not update for quite some time due to the heavy changes that have been made, until now where I found the API is up and running again and everything is good again with the new (old) Forge. Well done, respect for all your hard work =)

Here comes my question: by when do we expect the API to allow to send in the Fulx relevant parameters during txt2img? By now I can run flux models via the API but they come out broken most the time maybe due to wrong VAE or using the wrong clip/t5 or something I just dont see its missing.

Is there any date available? Or is it already possible and I did not see it in the FastAPI documentation?

If you wonder why I like the API so much, have a look at my baby https://github.com/osi1880vr/prompt_quill

Thanks for your time

[catto123455]

i am waiting too，i use this，and returned image is pure black. ：
import requests
import base64
from PIL import Image
from io import BytesIO
import os
import json # 导入 json 模块

使用的API接口基础URL

web_ui_url = "http://xxxxxx:7862" # 替换为你的服务地址和端口

构造 payload：文本到图像需要的参数

payload = {
"prompt": "1girl", # 正向提示词
"negative_prompt": "low quality, bad anatomy, text", # 反向提示词
"seed": -1, # 随机种子（-1 表示随机）
"steps": 20, # 采样步数
"cfg_scale": 1, # CFG Scale
"width": 896, # 图像宽度
"height": 1152, # 图像高度
"batch_size": 1, # 每批图像数量
"n_iter": 1, # 生成批次
"sampler_index": "Euler", # 采样方法
"schedule_type": "Simple", # 调度类型
"send_images": True, # 是否返回图片数据
"save_images": False, # 是否在服务端保存图像
"override_settings": {
"sd_model_checkpoint": "flux-dev-fp8.safetensors", # 模型名称
"sd_vae": "flux_ae.safetensors", # VAE 名称
}
}

向Stable Diffusion WebUI API发送Post请求以生成图片

response = requests.post(f"{web_ui_url}/sdapi/v1/txt2img", json=payload)

检查响应状态，成功时处理

if response.status_code == 200:
json_data = response.json() # 读取返回的JSON数据
images = json_data['images'] # 获取生成的图像Base64编码数组

# 指定保存路径
output_dir = r"E:\output"  # 保存图片的目录
os.makedirs(output_dir, exist_ok=True)  # 如果目录不存在，则创建目录

# 处理每张图片，将Base64图像解码，然后使用PIL保存/显示
for i, img_data in enumerate(images):
    img_bytes = base64.b64decode(img_data)  # 将Base64数据解码为字节流
    img = Image.open(BytesIO(img_bytes))  # 将字节流转换为PIL图像对象

    # 保存图片到指定目录
    output_path = os.path.join(output_dir, f"output_image_{i}.png")
    img.save(output_path)  # 保存图片
    print(f"图片已保存到: {output_path}")

    # 保存生成参数到文本文件
    parameters = json_data.get('parameters', {})  # 获取参数，默认为空字典
    with open(os.path.join(output_dir, f"output_image_{i}_parameters.txt"), "w", encoding="utf-8") as f:
        f.write(json.dumps(parameters, indent=4))  # 将字典转换为 JSON 字符串并写入文件

    img.show()  # 显示图片

else:
print(f"Error: {response.status_code}, {response.text}")

[jtabernik]

I was having the same issue. No idea if this will work for you or not, but I added these two parameters to my payload and it fixed it. I see you have scheduler and sampler specified already, but with different parameter names. Anyway, hope this helps!

    "sampler_name": "Euler",
    "scheduler": "Simple"