Metadata and Prompts

AIGC Metadata & Prompt Inspection

Berry AI Studio features a lossless, multi-engine metadata parser written in Rust (berry-metadata). It automatically extracts prompts, negative prompts, models, seeds, and execution graphs across all major AI generation platforms.

---

1. Supported AI Generation Platforms

Berry natively understands metadata embedded in PNG chunks, WebP EXIF headers, and MP4 ISOBMFF boxes from:

Platform / Tool	Extracted Metadata Fields	Source Container Location
AUTOMATIC1111 / SD.Next / Forge	Prompt, Negative Prompt, Steps, Sampler, CFG, Seed, Size, Model Hash, Model Name, Denoising, Hires Upscale	PNG parameters chunk / JPEG EXIF UserComment
ComfyUI	Full node execution graph, positive/negative CLIP text encode, KSampler seeds, steps, CFG, checkpoint loaders, LoRA loaders, latent upscale nodes	PNG prompt & workflow chunks / WebP ComfyUI chunks / MP4 moov/udta
NovelAI	Title, Description, Prompt, Negative Prompt, Seed, Sampler, Steps, Scale, Software version	PNG Comment & Description chunks
Fooocus / Fooocus-MRE	Base Model, Refiner, LoRA weights, Sharpness, Performance mode, Resolution, Prompt	PNG parameters text blocks
InvokeAI	Model name, VAE, Scheduler, Generation mode (txt2img/img2img), Seamless tiling	PNG sd-metadata & invokeai_metadata JSON
EasyDiffusion / Stable Swarm	Formatted JSON parameter blocks, Seed, Model name	PNG sui_image_params / sidecars

---

2. The Property Inspector Pane (InspectorPane.vue)

When an image or video is selected, the right-hand panel (I hotkey) presents its metadata cleanly:

┌────────────────────────────────────────────────────────┐
│ PROPERTY INSPECTOR                                 [✕] │
├────────────────────────────────────────────────────────┤
│ [ Media Thumbnail Preview ]                            │
│ 1024 × 1024 · PNG · 3.4 MB · 2026-09-21                │
│ [ ★★★★★ ]  [ ★ Favorite ]  [ 🔞 NSFW ]                │
├────────────────────────────────────────────────────────┤
│ Positive Prompt                           [📋 Copy]    │
│ ┌────────────────────────────────────────────────────┐ │
│ │ [masterpiece] [1girl] [solo] [cyberpunk city]      │ │
│ │ [neon reflections] [rain] [volumetric lighting]    │ │
│ └────────────────────────────────────────────────────┘ │
├────────────────────────────────────────────────────────┤
│ Negative Prompt                           [📋 Copy]    │
│ ┌────────────────────────────────────────────────────┐ │
│ │ worst quality, low quality, bad anatomy, bad hands │ │
│ └────────────────────────────────────────────────────┘ │
├────────────────────────────────────────────────────────┤
│ Generation Parameters                                  │
│ Model:    animagine_xl_3.1.safetensors                 │
│ Hash:     31e35c80  [🔍 Find in Civitai]               │
│ Sampler:  DPM++ 2M Karras                              │
│ Steps:    28             CFG Scale: 7.0                │
│ Seed:     2849104812     [📋 Copy]                     │
├────────────────────────────────────────────────────────┤
│ Detected LoRAs (2)                                     │
│ • CyberpunkStyle (Weight: 0.85)           [+ In Prompt]│
│ • DetailedEyes (Weight: 0.6)              [+ In Prompt]│
├────────────────────────────────────────────────────────┤
│ ▼ Raw Metadata (ComfyUI Workflow JSON)                 │
└────────────────────────────────────────────────────────┘

---

3. Interactive Prompt Token Chips

Berry AI Studio parses prompt strings into tokenized chips rather than displaying a wall of unformatted text:

• 1-Click Search: Clicking any token chip (e.g. [cyberpunk city]) immediately executes a library search for that specific concept across all your artworks.
• 1-Click Copy: Click the copy icon in the upper-right corner of the prompt box to copy the clean, unescaped prompt directly to your system clipboard.
• Tag Discovery: Tokenized chips make it easy to identify artist styles, lighting keywords, or quality tags that you want to reuse.

---

4. Checkpoint Models & Hash Identification

Generative tools often embed short model hashes (e.g. 31e35c80) or full SHA256 hashes instead of human-readable filenames.

• Berry automatically queries its local Model Cache (model_cache SQLite table) to map cryptic hashes into friendly checkpoint names like "Animagine XL 3.1".
• If an unknown hash is encountered, you can import an AUTOMATIC1111 cache.json file or look it up directly via Civitai (see Models & LoRA Library).

---

5. Raw Metadata & ComfyUI Workflow JSON

For advanced users and technical directors who need to examine node links:

• Expand the Raw Metadata accordion at the bottom of the inspector to inspect the unedited JSON payload.
• You can copy the entire JSON workflow block to paste directly into text editors or share with colleagues.