SiliconFlow

SiliconFlow (/media/siliconflow) is a fast OpenAI-compatible inference platform serving hundreds of open and commercial models. Its media studio covers image generation, Wan2.x video, and text-to-speech in one place.

Where to configure

SiliconFlow reuses your org's existing Settings → AI SiliconFlow key as a universal credential, so no separate Settings → Media entry is required. You can still add a dedicated media credential there if you prefer.

See Settings for provider setup.

Tabs

Tab	Operation	Key fields	Output
Text → Image	Image	Model (live catalog/curated list), prompt, negative prompt, size, steps, batch size, seed	Still image(s)
Text → Video	Video	Model (Wan2.x), prompt, negative prompt, resolution, seed	Video clip
Image → Video	Video	Model (Wan2.x), source image, prompt, resolution	Video clip
Text → Speech	Audio	Model (Fish-Speech / CosyVoice), text, voice, format (MP3/WAV)	Audio file

Generation flow

The studio follows the standard Studio Kit flow: fill the form, click Generate, track the job in the Render Queue, and open the finished asset in the Designer or the composer. See Media Studios overview for details.

Caveats

• Image and Text → Speech are synchronous — results appear almost immediately.
• Video is async and has no webhook; completion is driven by the media-jobs poll cron.
• Image-to-video base64-encodes the source frame before sending it to SiliconFlow.
• Image models are discovered from the provider catalog; video/audio models use a curated fallback list.

---

Verified against v1.0.0