Model Playground — ZorgDeepCore

The Model Playground is a chat-style interface for running AI generation directly on your GPU pod — without leaving the platform. It supports image generation, single-avatar talking video, dual-avatar talking video, and audio-driven video via LTX-2.3.

> 📸 [Screenshot: DeepCore Playground interface]

Getting to the Playground

1. Open GPU Pods from the sidebar.
2. Click Details on a running pod whose ComfyUI is ready.
3. Click Try Models in the pod detail header.

The playground is available at /gpu-pods/[instanceId]/playground. The Try Models button is only active when the pod is running and ComfyUI has finished starting up.

> 📸 [Screenshot: Try Models button on pod detail page]

Interface Layout

The playground is a full-screen chat interface:

• Top header — pod ID, connection status badge ("Connected"), and a back arrow to return to the pod detail page.
• Scrollable results feed — generated images and videos appear in a chat-bubble layout. Your prompt appears on the right; the result appears on the left with a DeepCore label and mode badge.
• Bottom input bar — mode selector, settings, file attachments, prompt field, and Generate button.

Each result includes:

• The generated image or inline video player.
• A timestamp.
• A Download button.

Generation Modes

Select the mode using the dropdown in the bottom input bar. Switching modes clears all current inputs.

Image Generation (Z-Image Turbo)

Generates images from a text prompt using the Z-Image Turbo model.

Inputs:

• Prompt — describe what you want to generate. Press Enter or click the Send button to generate Shift+Enter for a new line).
• Aspect ratio — choose from:

Output: An image displayed inline. Download with the button below the result.

> 📸 [Screenshot: Image mode with aspect ratio selector]

Single Avatar (InfiniteTalk)

Animates a still image to speak in sync with an audio track using InfiniteTalk.

Inputs:

• Source Image — upload an image file (jpg/png) or select from your asset repository. A preview thumbnail appears as a chip once uploaded.
• Audio — upload an audio file. The avatar will lip-sync to this audio.
• Prompt (optional) — describes the scene context (default: "a person is talking").

Output: A 480 × 832 video with the avatar speaking. Plays inline.

> 📸 [Screenshot: Single avatar mode with image and audio chips]

Dual Avatar (InfiniteTalk Multi)

Animates a still image containing two people, each speaking from a separate audio track.

Inputs:

• Source Image — an image with two visible people.
• Left Audio (blue chip) — audio for the left person.
• Right Audio (green chip) — audio for the right person.
• Prompt (optional) — describes the scene (default: "the woman and the man are talking").

Output: A 480 × 832 video with both avatars speaking their respective audio tracks.

> 📸 [Screenshot: Dual avatar mode with left/right audio chips]

LTX-2.3 Video (Audio-to-Video)

Generates a video from an image and an audio file using LTX-2.3. This mode creates cinematic video driven by both the image and audio.

Inputs:

• Source Image — select from your Asset Library (opens the Asset Manager modal).
• Audio — select from your Asset Library. If you select an audio file, the duration field is auto-populated with the audio length.
• Description — main description of the video content.
• Scene (optional) — scene details (e.g. "outdoor cafe, sunny day").
• Character (optional) — character description.
• Action (optional) — what the character is doing.
• Camera (optional) — camera movement or framing (e.g. "slow zoom in").

Video settings (shown in the toolbar):

• Width — 256–1920 px (default 1280, step 64).
• Height — 256–1920 px (default 720, step 64).
• Duration — 1–30 seconds (default 9; auto-set from audio length when available).

The description and optional fields are combined into a structured prompt sent to the model.

Output: A video with the specified dimensions and duration. Plays inline.

> 📸 [Screenshot: LTX-2.3 mode with structured prompt fields and video settings]

File Uploads

Files are uploaded to cloud storage before generation begins. While a file is uploading, a spinner appears on the chip and the Generate button is disabled.

• Image — any standard image format (jpg, png, webp, etc.).
• Audio — any standard audio format (mp3, wav, m4a, etc.).

For Single Avatar and Dual Avatar modes, files are uploaded directly from your device via the chip button.

For LTX-2.3 mode, images and audio are picked from the Asset Library (Media Jobs output and repository assets) rather than uploaded from disk.

Generating

Click the Generate button (send icon) or press Enter (in single-line modes) to start generation. While generating:

• The send button shows a spinner.
• A status message appears in the feed (e.g. "Submitting job…", "Generating image…", "Rendering avatar video…").
• Results stream in automatically as each job completes (polled every 5 seconds).

You can queue multiple generations — results appear in order as they complete.