AI video and image generation at cost — your keys, real prices, no subscription
AI runs online
Prices are for one typical use: one request of about 1,500 words sent and 375 words back. A higher quality score is better.
Hickeyfield has no option to run its AI on your computer.
| Model | Quality | Per 1,000 uses |
|---|---|---|
| On your computer (Ollama, 4-bit download size) | ||
| Granite 4.2 3B2.2 GB | 9 | $0 |
| Phi-4 Mini2.5 GB | 6 | $0 |
| Llama 3.1 8B4.9 GB | 7 | $0 |
| gpt-oss 20B14 GB | 9 | $0 |
| Gemma 4 31B20 GB | 19* | $0 |
| Qwen3.5 35B-A3B24 GB | 19* | $0 |
| publik API | ||
| publik-fastGLM-5.3 Flash | 42 | $1.00 |
| publik-balancedMiMo-V2.6-Pro · Hickeyfield | 46 | $10.00 |
| publik-smartGPT-6 Sol | 48 | $18.00 |
Quality: Artificial Analysis Intelligence Index v4.3.2, read 2026-09-22 (publik API models 2026-09-25); * = estimated by Artificial Analysis. Sizes: the Ollama library, read 2026-09-22.
Every step written out. No terminal experience needed. Pick your setup.
A free, open-source desktop studio for AI video and image generation. Bring your own API keys and pay providers directly at cost — there is no subscription, no credits, no plan gating, and nothing to pay us.
A hickey is the speck of dirt that prints as a haloed blemish — the flaw that proves a thing went through a real press rather than a renderer.
Hickeyfield is an independent open-source project. It is not affiliated with, endorsed by, or sponsored by Higgsfield, Inc.
Status: early, and honest about it. Hickeyfield generates: you add a provider key, pick a use case, attach media if the job needs it, and get real video and images back, saved to a folder you choose. What is verified end to end is the fal route on macOS — upload, submit, poll, download, play. Windows builds in CI on every commit but has not been exercised by hand. The surfaces beyond the generator (library, audio, studios) are not built yet.
The best AI media tools are harnesses: they wrap third-party models in presets and a good interface. The interface is the product, and the interface is reproducible. What isn't reproducible is subsidised bulk model pricing — which is exactly what a subscription buys you, and exactly what you don't need if you already hold your own API keys.
So: same features, same workflow, your keys, at cost.
On flagship video you may pay slightly more per generation than a bulk subscriber, because large platforms buy in volume and subsidise. On Veo Fast/Lite, Grok direct, Gemini Omni, Seedream, Recraft, upscaling and every self-hosted path, you pay considerably less. And you never pay for a month you didn't use.
Hickeyfield always shows the real USD cost before you submit, computed from live provider price feeds rather than an opaque credit integer.
The generator is organised by use case rather than by model, and each one offers only the models that can actually do that job — so a model that cannot take a video never appears under Edit Video, instead of failing after you have attached one and pressed Generate.
Everything a request depends on is checked against the provider's own published schema before anything is sent: which inputs the endpoint accepts, which values it enumerates, and what shape the result will be. A setting the endpoint does not have is reported rather than silently dropped, and a value it will not accept is refused before you are charged rather than after.
Prompts can be expanded through a filmmaking corpus — camera, lens, light, motion — either deterministically from a preset, or through a local model via Ollama. Nothing is sent anywhere for this unless you ask for it.
Tauri v2 — a Rust core with a React UI in the system webview.
crates/hickeyfield-core/ provider adapters, routing, job engine, presets, costs
(no Tauri dependency, so tests run in seconds)
src-tauri/ the desktop shell: commands, events, window lifecycle
ui/ React + Vite frontend
scripts/ signing, verification and provenance tooling
provenance/ one-way hashes used by the copy-provenance lint
Being native rather than a web app is load-bearing, not incidental:
Two CI lints protect the line between reimplementing a workflow and copying someone's work:
The model catalogue is seeded from higgsfield-ai/cli's MODELS.md, which is
MIT licensed and published by its authors as a machine-readable specification.
See NOTICE.
AGPL-3.0-or-later. See LICENSE, and NOTICE for bundled third-party
components.