Pricing
FastGen Cloud is prepaid and usage-based. You add funds to your account balance by card (minimum $5, maximum $2,000 per top-up) and each job is charged from that balance at the rates below when it is submitted. Failed or blocked jobs are refunded to your balance automatically. There are no subscriptions and no minimum monthly commitment.
Per-job rates
| Operation | Model | Price (USD) |
|---|---|---|
| Text to image (≤ 1 MP) | krea-2, krea-2-anime, krea-2-cartoon | $0.006 / image |
| Image edit | qwen-image-edit | $0.03 / image |
| Face ID (image to image, ≤ 1 MP) | sdxl-face-id | $0.012 / image |
| Image to image, inpaint and character generations (≤ 1 MP) | krea-2, krea-2-anime, krea-2-cartoon | $0.006 / image |
| Image to video | wan-2.2 | $0.05 / second |
| Image to video with audio | minimax-h3 | $0.06 / second |
| Image to video with audio (hosted) | wan-3 | $0.09 / second |
| Character training | characters | $0.75 / training |
| Character slots — 10 | 10 characters (2 free on every account) | $10.00 / month |
| Character slots — 100 | 100 characters | $39.00 / month |
| Character slots — 500 | 500 characters | $129.00 / month |
- Images larger than 1 megapixel are billed pro rata by pixel count (e.g. a 1.5 MP image costs 1.5×).
- Video is billed per second of output; 5-second and 8-second clips are available.
- Prices are in US dollars and exclude any applicable taxes.
Endpoints
Krea 2
krea-2
Krea 2 (12B) photorealistic text-to-image. Write prompts as plain sentences — subject, action, clothing, setting, light, camera. Bring your own Krea 2 LoRAs by URL. Image-to-image, inpainting and trained-character generations run on a photoreal SDXL companion under this same model id (SDXL LoRAs there).
License: Krea 2 Community License (commercial use under $1M annual revenue)
SDXL Face ID
sdxl-face-id
Identity-preserving image-to-image on SDXL (IP-Adapter FaceID): one reference photo of a person + a prompt → new images of the same person in the described scene. style: realistic (photoreal, default), anime, cartoon — likeness is strongest on realistic. Bring your own SDXL LoRAs by URL (match the style's checkpoint family).
License: Open-weight (IP-Adapter Apache-2.0; checkpoint licences per model card)
Qwen Image Edit
qwen-image-edit
Qwen-Image-Edit instruction-based editing from 1–3 reference images: change objects, poses, styles or backgrounds, or combine subjects. Bring your own Qwen-Image-Edit LoRA by URL (one, next to the built-in anatomy LoRA).
License: Apache-2.0
Krea 2 Anime
krea-2-anime
Krea 2 (12B) with a dedicated anime style: anime and illustration text-to-image from plain-sentence prompts. Bring your own Krea 2 LoRAs by URL. Image-to-image, inpainting and trained-character generations run on an Illustrious SDXL companion under this same model id (Illustrious LoRAs there).
License: Krea 2 Community License (commercial use under $1M annual revenue)
Krea 2 Cartoon
krea-2-cartoon
Krea 2 (12B) comics / cartoon finetune: bold line work, flat colour, expressive characters, from plain-sentence prompts. Bring your own Krea 2 LoRAs by URL. Image-to-image, inpainting and trained-character generations run on a Pony SDXL companion under this same model id (Pony LoRAs there).
License: Krea 2 Community License (commercial use under $1M annual revenue)
Wan 2.2
wan-2.2
Wan 2.2 image-to-video: animate a still image into a 5 or 8 second clip at 32 fps; the start frame's aspect ratio is preserved. Bring your own WAN 2.2 LoRAs by URL (high-noise `url` + optional low-noise `low_url`).
License: Apache-2.0
MiniMax H3
minimax-h3
MiniMax H3 (33B omni-modal) image-to-video with native audio: a start frame + a prompt → a 5, 8 or 10 second clip at 24 fps with a generated soundtrack. The canvas follows the start frame (768 px short edge). Bring your own MiniMax H3 LoRAs by URL (up to 3). No negative prompt.
License: MiniMax H3 Community License (open weights; the licence excludes the USA, EU, UK and Korea from its applicable territory — production use there needs a separately granted licence from MiniMax)
Wan 3
wan-3
Wan 3.0 Prime image-to-video (hosted): a start frame + a prompt → a 5–15 second clip with native audio at 480p, 720p or 1080p. Prompt-only — no LoRAs, no negative prompt. Longer to render than wan-2.2; higher fidelity and sound.
License: Hosted model (Atlas Cloud); Wan 3.0 Prime terms apply
Billing details
- Payment methods. Major credit and debit cards, processed by Stripe. We never store card numbers.
- Receipts and invoices. Every top-up produces a receipt and a downloadable invoice.
- Auto-recharge. Optionally keep a card on file and automatically top up when your balance falls below a threshold you choose.
- Refunds. Unused prepaid balance is refundable on request within 30 days of purchase, less payment-processing fees. See the Refund & Cancellation Policy.
- Volume. Generating more than $2,000 / month? Contact us for volume pricing and dedicated capacity.