freegen-image-generation

Freegen Image Generation

Übersicht

freegen ist ein Docker-Container auf dem homedocker-Server, der die Dezgo.com API wrapped.
Interne URL: http://freegen:3000

Alle Generierungen laufen über einen einzigen Endpunkt: POST /generate
Bild-Download: GET /files/{filename}.png
Info: GET /info
Health: GET /health

---

Endpoints

POST /generate

Der zentrale Endpoint. Unterstützt folgende category-Werte:

---

Payload-Formate

TXT2IMG

Flux (zuverlässig, schnell, uncensored):

{
  "prompt": "a red dragon breathing fire, fantasy art, detailed",
  "category": "flux",
  "width": 1024,
  "height": 1024,
  "steps": 4
}

SDXL:

{
  "prompt": "a portrait of a young woman with purple hair, photorealistic",
  "category": "sdxl",
  "model": "dreamshaperxl_1024px",
  "width": 768,
  "height": 1024,
  "guidance": 7,
  "steps": 30,
  "seed": 2024
}

SD 1.5:

{
  "prompt": "a cute cat on a windowsill, detailed fur",
  "category": "sd1",
  "model": "realistic_vision_5_1",
  "width": 512,
  "height": 768,
  "guidance": 7,
  "steps": 30,
  "sampler": "dpmpp_2m_karras",
  "seed": 2024
}

SDXL Lightning:

{
  "prompt": "a cyberpunk cityscape at night, neon lights",
  "category": "sdxl2",
  "model": "juggernautxl_10_hyper",
  "width": 768,
  "height": 1024,
  "guidance": 2,
  "steps": 10
}

IMG2IMG (Image-to-Image) — Outfit-Änderung, Style-Transfer

{
  "prompt": "person wearing a red summer dress",
  "category": "i2i",
  "model": "realistic_vision_5_1",
  "init_image": "data:image/png;base64,...",
  "strength": 0.75,
  "width": 512,
  "height": 768,
  "steps": 30
}

Inpainting — Bereich ersetzen

Mit manueller Maske (SD 1.5):

{
  "prompt": "a silver belly button piercing",
  "category": "inpaint",
  "init_image": "data:image/png;base64,...",
  "mask_image": "data:image/png;base64,...",
  "strength": 1,
  "steps": 30,
  "sampler": "euler"
}

Mit manueller Maske (SDXL):

{
  "prompt": "a silver belly button piercing",
  "category": "inpaint_sdxl",
  "init_image": "data:image/png;base64,...",
  "mask_image": "data:image/png;base64,...",
  "strength": 1,
  "steps": 20
}

Per Text-Prompt maskieren (✨ am einfachsten!):

{
  "prompt": "a belly button piercing with a silver ring",
  "category": "textinpaint",
  "mask_prompt": "belly button, navel",
  "init_image": "data:image/png;base64,...",
  "strength": 0.85,
  "steps": 20,
  "sampler": "euler"
}

ControlNet — Struktur-gesteuert

{
  "prompt": "a person in the same pose, wearing armor",
  "category": "controlnet",
  "init_image": "data:image/png;base64,...",
  "control_model": "canny",
  "control_scale": 1
}

control_model: canny, hed, openpose, depth, normal

Remove Background

{
  "category": "removebg",
  "init_image": "data:image/png;base64,..."
}

Upscale

{
  "category": "upscale",
  "init_image": "data:image/png;base64,...",
  "scale": 2
}

scale: 2, 3, oder 4

---

⚠️ Hinweise

Dezgo Content Policy (Stand 28.07.2026):

Workaround für restriktive Prompts:

• Flux TXT2IMG mit direktem Prompt (kein Filter)
• Clean-Inhalte funktionieren mit allen Tools einwandfrei (Kleidungswechsel, Piercings, Styling etc.)

---

Status-Tabelle

---

Beispiel-Workflow (Node.js/OpenClaw)

// Bild generieren
const res = await fetch('http://freegen:3000/generate', {
  method: 'POST',
  headers: { 'Content-Type': 'application/json' },
  body: JSON.stringify({
    prompt: 'a cute orange cat',
    model: 'flux'
  })
});
const data = await res.json();
// data = { success, file, url, size, elapsed, cost, balance, model, sub_model, params }

// Bild runterladen
const imgRes = await fetch(`http://freegen:3000${data.url}`);
const imgBuf = Buffer.from(await imgRes.arrayBuffer());

---

Verfügbare Modelle

Komplette Liste via GET /info abrufbar.

SDXL: dreamshaperxl_1024px, juggernautxl_1024px, ponyxl_6, realismenginesdxl_1024px, sdxl_1024px, bluepencilxl_1024px, illustrious_*

SDXL Lightning (sdxl2): juggernautxl_10_hyper, dreamshaperxl_lightning_1024px, ponyxl_6_turbo_dpo, envy_starlight_xl_01_lightning_1024px

SD 1.5 (sd1): realistic_vision_5_1, deliberate_2, juggernaut_reborn, dreamshaper_8, absolute_reality_1_8_1 + 99+ weitere

---

Fehlerbehandlung

• Flux: model und category können fehlen (defaults to flux)
• SDXL/SD1.5/SD2: category + model sind beide nötig
• HTTP 400 InvalidEnumValue: Falscher Model-Name
• **HTTP 499: Timeout/Abort – Request war zu langsam oder zu groß
• Fake Images (310 bytes): Flux 2, Flux 2 Pro, Grok Imagine, Nano Banana sind Broken Models – nicht verwenden!