Skip to content

img-forge

11 models. One endpoint.

Serverless image generation and iterative editing — built for autonomous agents, proven in real production pipelines. Every visual asset across our own products, from game art to social posts, runs through this API. Monthly quota in every Stackbilder plan; Agency volume when the pipeline scales.

Powered by Black Forest Labs Leonardo Stability AI ByteDance Lykon RunwayML Google DeepMind (coming soon)

Same prompt. Every model.

Eight perspectives. One endpoint.

“floor-to-ceiling office windows at golden hour, empty standing desk silhouetted against the city below, warm and cool light split, editorial architectural photography”

floor-to-ceiling office windows at golden hour, empty standing desk silhouetted against the city below, warm and cool light split, editorial architectural photography — FLUX 2 Klein 4B by Black Forest Labs

FLUX 2 Klein 4B

Black Forest Labs

floor-to-ceiling office windows at golden hour, empty standing desk silhouetted against the city below, warm and cool light split, editorial architectural photography — FLUX 2 Klein 9B by Black Forest Labs

FLUX 2 Klein 9B

Black Forest Labs

floor-to-ceiling office windows at golden hour, empty standing desk silhouetted against the city below, warm and cool light split, editorial architectural photography — FLUX 1 Schnell by Black Forest Labs

FLUX 1 Schnell

Black Forest Labs

floor-to-ceiling office windows at golden hour, empty standing desk silhouetted against the city below, warm and cool light split, editorial architectural photography — Lucid Origin by Leonardo

Lucid Origin

Leonardo

floor-to-ceiling office windows at golden hour, empty standing desk silhouetted against the city below, warm and cool light split, editorial architectural photography — Phoenix 1.0 by Leonardo

Phoenix 1.0

Leonardo

floor-to-ceiling office windows at golden hour, empty standing desk silhouetted against the city below, warm and cool light split, editorial architectural photography — SDXL Lightning by ByteDance

SDXL Lightning

ByteDance

floor-to-ceiling office windows at golden hour, empty standing desk silhouetted against the city below, warm and cool light split, editorial architectural photography — DreamShaper 8 by Lykon

DreamShaper 8

Lykon

floor-to-ceiling office windows at golden hour, empty standing desk silhouetted against the city below, warm and cool light split, editorial architectural photography — SDXL Base 1.0 by Stability AI

SDXL Base 1.0

Stability AI

Quality tiers

Three tiers. Nine models. One auth.

Switch tiers per-request — no integration rewrite, no per-provider account setup. Pin a specific model with cf_model or let img-forge route to the tier default.

Draft ~4s · 1 credit

Fast prototyping. 4-step distillation.

Rapid iteration. Concept exploration. Moodboards. Agent loops that need fast signal.

SDXL Lightning default
FLUX Schnell FLUX family speed tier
DreamShaper 8 LCM Stylized output
Standard ~10s · 2 credits

Daily-driver quality. 25-step FLUX.

Blog headers. Product shots. Content pipelines. The tier most production agents use by default.

FLUX Klein 4B default
FLUX Klein 9B Larger model, same cost
SDXL Base 1.0 Full SDXL, no distillation
Premium ~20s · 5 credits

Portfolio quality. High detail and creative control.

Marketing hero art. Editorial. Architectural visualizations. Outputs worth showing clients.

FLUX Dev default
Lucid Origin Photorealistic, 1120px native
Phoenix 1.0 High creativity, neg. prompt support

Not coming soon. Community-funded.

We don't lock models to look exclusive. Frontier image models cost real money every single time they run. We turn them on when the community has prepaid the keys — and we show you the live backer count. Hit the number, Pro subscribers get access.

Ultra Pro required Community unlock

Gemini 3.1 Flash · Google DeepMind

2K native · 10 credits

0 backers 100 to unlock

Funded, not promised. Pro subscribers unlock when we hit the number. Every pack moves the bar.

→ Get Pro
Ultra+ Pro required Community unlock

Gemini 3 Pro · Google DeepMind

4K native · 20 credits

0 backers 100 to unlock

Funded, not promised. Pro subscribers unlock when we hit the number. Every pack moves the bar.

→ Get Pro
OpenAI direct models The openai_model API parameter routes requests to OpenAI's image generation models directly: gpt-image-1-mini, gpt-image-1, gpt-image-1.5, and gpt-image-2. These consume ultra-tier quota and billing — the same entitlement checks, quota guards, and credit accounting apply. Access requires Stackbilt-managed OpenAI provider credentials; you do not supply your own OpenAI key. Available to ultra-tier accounts; other tiers receive a 403. gpt-image-1 — available now; OpenAI retires it Oct 23, 2026. Listed so existing pipelines aren't surprised. New builds: target gpt-image-1.5 or gpt-image-2. gpt-image-1.5 and gpt-image-2 are community-milestone-locked — see the tiers above for unlock status.

Iterate & refine

Generate. Edit. Correct. One endpoint.

img-forge isn't generate-only. Two editing models let agents build iterative pipelines — restyle an image or surgically fix the 10% that's wrong. Available via REST API and two MCP endpoints: a dedicated img-forge MCP (imgforge-mcp.stackbilder.com) for agents that only need image tools, or the Stackbilder MCP gateway for the full platform.

Source: abandoned greenhouse, photorealistic photography
Source
Output: oil painting restyle, impressionist style
strength: 0.72
Output
img2img stable-diffusion-v1-5-img2img

Style transfer & quality uplift

Send a source image alongside your prompt. Chain a draft generation directly into a high-quality restyle — pass input_asset_url from a prior call and skip the base64 round-trip entirely.

// image_generate MCP tool — chain from a prior result
image_generate({
  prompt: "photorealistic DSLR quality, ultra detailed",
  cf_model: "@cf/runwayml/stable-diffusion-v1-5-img2img",
  input_asset_url: prev.asset_url,  // no base64 fetch needed
  strength: 0.75
})
Source: product shot with artifact region to correct
Source + mask artifact region
Output: inpainted result, artifact corrected
corrected
Output
inpainting stable-diffusion-v1-5-inpainting

Fix the 10% that's wrong

Asset 43 of 67 has a bad hand. Mask the region, fix it, keep everything else — no re-roll, no drift from the approved look. In agent pipelines, a vision model detects the failure and runs the same call unattended, at draft-tier cost.

// image_generate MCP tool — surgical region fix
image_generate({
  prompt: "clean background, no artifacts",
  cf_model: "@cf/runwayml/stable-diffusion-v1-5-inpainting",
  input_asset_url: sourceResult.asset_url,  // chain from prior call
  mask_image: maskBase64   // white = regenerate, black = keep
})

Available via REST API and MCP · mcp.stackbilt.dev/mcp

The API

One request. One URL back.

POST a prompt, get an asset_url back. Async jobs for premium tier; synchronous for draft and standard.

# Generate — standard tier (default)
curl -X POST https://imgforge.stackbilt.dev/v2/generate \
  -H "Authorization: Bearer sb_live_..." \
  -H "Content-Type: application/json" \
  -d '{
    "prompt": "a red fox in a snowy forest, golden hour",
    "quality_tier": "standard"
  }'

# Response
{
  "job_id": "01J...",
  "state": "completed",
  "asset_url": "/v2/assets/7082f9...",
  "quality_tier": "standard"
}

# Pin a specific model
curl -X POST https://imgforge.stackbilt.dev/v2/generate \
  -H "Authorization: Bearer sb_live_..." \
  -H "Content-Type: application/json" \
  -d '{
    "prompt": "portfolio-grade hero art, dramatic lighting",
    "cf_model": "@cf/leonardo/phoenix-1.0"
  }'

The design workflow

From napkin to pixel-perfect. One API.

Wireframe the idea. Board the brand. Mock the launch page. Every image below is img-forge output — from the same pipeline that produces the game art for The Tarot Detective, the content assets for Stack Arcade, and the design direction for this page. Our agent runs the calls; the creative direction is ours. Three calls, one API — the same three your agent or your pipeline would make.

Hand-drawn lo-fi wireframe sketch of a SaaS analytics dashboard, pencil style with annotated margin notes 01

Wireframe

ultra

Lo-fi annotated wireframe, SaaS dashboard

Brand-direction moodboard collage for a developer-tools startup showing typography, color palette, and UI component swatches 02

Brand direction

ultra

Moodboard, developer-tools startup

High-fidelity dark-themed landing page mockup for a developer API product, styled in the manner of Stripe or Vercel 03

Hi-fi mockup

ultra+

Dark landing page mockup, API product

Production proof

One game. 67 assets. One visual universe.

The Tarot Detective is a live detective game whose entire art pipeline runs on img-forge: 16 character portraits, 23 location backgrounds, 28 props and evidence shots. Every character keeps the same face across every scene. The workflow: two style systems — one for portraits, one for environments and props — plus a lock-after-approval rule. Once a character's look is approved, every subsequent call reuses the same seed — supported on draft, standard, and premium tiers — so asset 43 still matches asset 2.

Vera Saint-Claire character portrait, TAROT_NOIR style, locked seed Dr. Helena Price character portrait, TAROT_NOIR style, locked seed Vivian Delacroix character portrait, TAROT_NOIR style, locked seed
Assets in production 67
Total generation budget $15–25
Same list, human-illustration rates $3,350–13,400
Style systems, zero drift 2
Iron District location background, NOIR_DECO style — generated environment art for The Tarot Detective
Location background · NOIR_DECO
Gilded circle pin, game evidence prop Grandmother's tarot deck, game evidence prop The ledger, game evidence prop

Evidence props · SQUARE_1K

Stack Arcade, our gaming news hub, runs its content assets through the same pipeline. This is the workload the Agency tier exists for — see Agency pricing.

What you get

Flat pricing. No markup. No per-provider setup.

Included in your plan

No per-image fees. No token accounting. No surprise invoices. Monthly quota is flat and the same regardless of which model you use. Top up with credit packs when you need more.

11 models. One integration.

Switch between SDXL, FLUX, Leonardo, and Gemini per-request — no extra accounts, no per-provider SDK. One auth token, one billing surface, one endpoint. Agents don't need to know which model is running.

Edge-native & MCP-ready

Runs inside Cloudflare's network. Service-bound to your Stackbilder account. Generated images stored in R2 and retained for 30 days — DELETE /v2/jobs/:id removes the job record and R2 object immediately. MCP tool at mcp.stackbilt.dev/mcp for direct agent consumption.

Your agent can pay its own bills.

billing_status and billing_purchase_credits are MCP tools — an agent can check its balance mid-run, top up, and keep generating without a human touching a dashboard. Replicate, fal.ai, and Leonardo expose nothing like this to agents.

Quotas

Generous caps. No overage fees.

If you hit your cap, generation pauses until next cycle. No surprise charges, ever.

Free

$0

5

images / month

Draft and standard tiers. All 9 Cloudflare models available.

Pro

$29/mo

1,000

credits / month

All tiers including Ultra+. 100 ultra images or 200 premium. Credit packs stack on top.

Agency

$75/mo

4,000

credits / month · rollover

Built for production pipelines — a full game's asset run, a content calendar, client work at volume. Credits roll over while active. Priority queue and dedicated support.

Credits

Pay per image. Balance never expires.

Volume credits for FLUX and SDXL tiers — draft, standard, and premium. One-time packs carry forward indefinitely. Ultra and OpenAI models require a Pro subscription.

Every model's exact cost, before you spend a credit.

1cr draft → 20cr ultra+. No token math. No surprise invoice.

Draft1 cr
Standard2 cr
Premium5 cr
Ultra10 cr
Ultra+20 cr

Builder

500 credits

$12.50

one-time

  • · 500 draft images, or
  • · 250 standard images, or
  • · 100 premium images

Studio

Best value

1,000 credits

$20.00

one-time

  • · 1,000 draft images, or
  • · 500 standard images, or
  • · 200 premium images

Agency

Monthly

4,000 credits / mo

$75

per month · rollover credits

  • · A full production run (~350–500 credits) — eight per month, with room to spare
  • · Credits roll over — never reset while active
  • · 800 premium images / mo
  • · Per-model credit floor applies regardless of pack source

Builder + Studio credits never expire · Agency credits roll over · MCP-native via mcp.stackbilt.dev/mcp · Secured by Stripe

Start generating today.

Free tier includes 5 images across all Cloudflare models. No credit card.

or see how a full game got built →