Replicate

Replicate (/media/replicate) is a bespoke, native generative media workspace — not built on the generic Studio Kit — that runs models on Replicate using your own API token. It covers 19 generation categories across image, video, and audio in one page, plus a speech-to-text mode and two native editors (mask painting, video merge) and a canvas meme tool.

Categories

Category	Medium	Execution	Typical use
Text to Image	Image	Sync	Generate photos, illustrations, or concept art from a prompt.
Image to Image	Image	Sync	Transform or restyle an existing image.
Remove Background	Image	Sync	Remove the background of an image.
Upscale	Image	Sync	Increase resolution of an existing image.
Restore	Image	Async	Restore an old or degraded photo.
Inpaint	Image	Sync	Edit a painted region of an image with a mask.
Meme Generator	Image	Local (canvas)	Add draggable text layers to an image and export a PNG.
Text to Video	Video	Async	Generate a video from a prompt.
Image to Video	Video	Async	Animate an existing image into a video.
Video to Video	Video	Async	Transform an existing video with a style or effect.
Video Upscale	Video	Async	Increase resolution of an existing video.
Video Background	Video	Async	Remove/replace a video's background (matting).
Caption Video	Video	Async	Add burned-in captions/subtitles to a video.
Merge Videos	Video	Local (server-side ffmpeg)	Combine up to 6 clips with transitions into one MP4.
Text to Speech	Audio	Async	Convert text to spoken audio.
Text to Music	Audio	Async	Generate music from a prompt.
Music to Music	Audio	Async	Transform or extend an existing music clip.
Voice Clone	Audio	Async	Clone a voice from a short audio sample.
Speech to Text	Audio → Text	Sync	Transcribe an audio file to text.

Every run uses the org's own Replicate token, configured in Settings → Media. The category and model allowlist is curated server-side — Replicate hosts far more models than are exposed here.

---

Warm vs community models and cost badges

Each category lists warm official models by default. Warm models are pre-loaded by Replicate and start faster; the cost shown in the badge is fixed per run.

• Warm models are output-priced — the cost bar shows a fixed estimate before you click Generate.
• Community models are priced by compute time. Turn on the Show community models toggle to see them; they may cold-start, and the cost bar shows a usage-based estimate instead.

Select a model to reveal its dynamic input form, generated from the model's own schema. Required fields are marked; optional fields can be left at their defaults.

---

Save-folder requirement for async jobs

Every category marked Async above runs in the background. Before generating, choose a Save folder in the Files library — the finished asset imports into that folder automatically when the job completes. If no folder is selected, Generate stays disabled.

Sync categories (image generations, plus Speech to Text) complete immediately. If Replicate ends up queuing a sync prediction anyway, the studio silently falls back to registering an async job and polling until completion — the same save-folder rule applies.

---

Speech to Text — Copy + Download only

Speech to Text runs synchronously and returns the transcript inline. You can:

• Copy the full text to the clipboard.
• Download the transcript as .txt or .srt.

Transcripts are not saved to Files and do not consume media credits.

---

Inpaint mask painter

The inpaint workflow has two inputs:

1. Source image — pick an image from Files or provide a public https URL.
2. Mask — paint over the area you want to change with the brush tool. Use the eraser to refine. The exported mask is a black-and-white PNG where white pixels are the region to regenerate.

Adjust the brush size, then click Use this mask to return to the generation form. The source and mask are sent together to the selected inpainting model.

---

Video merge

The merge editor combines up to 6 clips into one MP4:

1. Click Add clip and choose a video from Files or paste a public https URL.
2. Re-order clips by dragging the list items.
3. Pick a transition and duration for each gap between clips.
4. Select a save folder and click Merge.

The job runs server-side via ffmpeg. When it completes, the merged video appears in the selected Files folder and can be opened directly from the result panel.

---

Meme generator

The meme editor is a native canvas tool:

1. Pick a base image from Files or a public https URL.
2. Add draggable text layers, each with its own font, size, color, outline, and style (bold/italic).
3. Drag text to position it on the canvas.
4. Click Export to save the meme as a PNG to Files, then Open in Designer to refine it further.

---

Related docs

• Media Studios — the shared render-queue/hand-off flow used by other studios.
• Media Library — folder tree, uploads, and storage routing.
• AI Tools — text generation, brand voice, and spend controls.
• Settings — configuring the Replicate media provider.

---

Verified against v1.0.0