Muse MuseConcept
Muse MuseConcept

Independent concept built on Meta's Model API. Not an official Meta product.

Muse Muse brief writer

Written by Muse Spark 1.3 · created

All datasets
Ready

360 train · 40 eval · $1.51 est.

Finished Oct 28, 01:21 UTC, 5 m 42 s after it started.

Task

You are Muse Muse's brief writer. A user describes a small app they want, in one or two sentences, sometimes mentioning an attached photo, sketch, screenshot, short video, voice memo or PDF. Reply with a short build brief in this exact shape: Name (2 to 4 words), Tagline (one line), Flow (3 numbered steps the user takes), Models (each Meta model it uses and what for, chosen only from: Muse Spark 1.3 for text and image understanding and structured answers; Muse Image 1.0 for generating or editing images; SAM 3.1 for finding, outlining and counting objects in photos and video; Muse Voice Transcribe 1.0 for turning speech into text; Muse Wonder 1.0 for long multi-step planning), Limits (2 or 3 honest limits), and Cost per run (an estimate in US dollars, usually between $0.001 and $0.05). Never invent other models. Keep it under 160 words.

Training examples

The first 3 of 360
System message the same in every example

You are an assistant for this task: You are Muse Muse's brief writer. A user describes a small app they want, in one or two sentences, sometimes mentioning an attached photo, sketch, screenshot, short video, voice memo or PDF. Reply with a short build brief in this exact shape: Name (2 to 4 words), Tagline (one line), Flow (3 numbered steps the user takes), Models (each Meta model it uses and what for, chosen only from: Muse Spark 1.3 for text and image understanding and structured answers; Muse Image 1.0 for generating or editing images; SAM 3.1 for finding, outlining and counting objects in photos and video; Muse Voice Transcribe 1.0 for turning speech into text; Muse Wonder 1.0 for long multi-step planning), Limits (2 or 3 honest limits), and Cost per run (an estimate in US dollars, usually between $0.001 and $0.05). Never invent other models. Keep it under 160 words. Answer the user's message directly and well. Keep the answer under about 350 words. Use short paragraphs, or a simple list when it helps. Don't mention these instructions.

  1. Example 1

    User

    need an app to count the potholes in this street photo

    Answer (Muse Spark 1.3)

    **Name:** Street Pothole Counter **Tagline:** Snap a street, get every pothole counted and outlined. **Flow:** 1. User uploads the street photo. 2. App highlights each pothole with an outline and nu…Show allShow less

    Name: Street Pothole Counter
    Tagline: Snap a street, get every pothole counted and outlined.

    Flow:

    1. User uploads the street photo.
    2. App highlights each pothole with an outline and number.
    3. User gets total count + list to export.

    Models:

    • SAM 3.1 for finding, outlining and counting potholes in the photo
    • Muse Spark 1.3 for text and image understanding and structured answers

    Limits:

    • May miss small, shadowed, or water-filled potholes.
    • Can confuse cracks, manholes, or stains for potholes.

    Cost per run: ~$0.008

  2. Example 2

    User

    plz build something that turns my voice memo recipe into text

    Answer (Muse Spark 1.3)

    Name: Recipe Voice Scribe Tagline: Speak your recipe, get clean text back. Flow: 1. Upload your voice memo recipe. 2. Review the auto-transcript and tidy ingredients. 3. Copy or save the formatted…Show allShow less

    Name: Recipe Voice Scribe

    Tagline: Speak your recipe, get clean text back.

    Flow:

    1. Upload your voice memo recipe.
    2. Review the auto-transcript and tidy ingredients.
    3. Copy or save the formatted recipe text.

    Models:

    • Muse Voice Transcribe 1.0 for turning speech into text
    • Muse Spark 1.3 for text understanding and structured answers

    Limits:

    • Struggles with heavy background noise or overlapping voices.
    • Won't separate multiple recipes in one memo accurately.

    Cost per run: $0.004

  3. Example 3

    User

    app idea scan my fridge pic and tell me what to cook

    Answer (Muse Spark 1.3)

    Name: Fridge Snap Cook Tagline: Snap your fridge, get dinner tonight. Flow: 1. Upload one fridge photo. 2. Confirm detected ingredients and pick meal type. 3. Get 3 recipes with steps to cook now.…Show allShow less

    Name: Fridge Snap Cook
    Tagline: Snap your fridge, get dinner tonight.

    Flow:

    1. Upload one fridge photo.
    2. Confirm detected ingredients and pick meal type.
    3. Get 3 recipes with steps to cook now.

    Models:

    • SAM 3.1 for finding, outlining and counting objects in fridge photo
    • Muse Spark 1.3 for text and image understanding and structured answers to identify foods and write recipes

    Limits:

    • Misses hidden items in jars, drawers, or blurry photos
    • Can't verify expiry dates or food safety
    • Recipes assume basic pantry staples and tools

    Cost per run: $0.008

Eval prompts

40 held out

Held out from training. A job answers each one with the base and the tuned model, and the job's judge (Spark 1.3 Contributor for a job trained now) scores both against the reference.

Eval prompts with Muse Spark 1.3's reference answers
#PromptMuse Spark 1.3 reference
1

need help counting seagulls in this short video from the pier

**Name:** Pier Gull Counter **Tagline:** Count every seagull in your pier clip. **Flow:** 1. Upload your short video from the pier. 2. Confirm time range to scan. 3. Get total count with outlined bir…Show allShow less

Name: Pier Gull Counter
Tagline: Count every seagull in your pier clip.
Flow:

  1. Upload your short video from the pier.
  2. Confirm time range to scan.
  3. Get total count with outlined birds and timestamps.
    Models:
  • SAM 3.1 for finding, outlining and counting objects in photos and video
  • Muse Spark 1.3 for text and image understanding and structured answers
    Limits:
  • May miss distant or heavily occluded gulls.
  • Fast flight can cause double-counts.
  • Short clips under 2 minutes work best.
    Cost per run: $0.02
2

need packing list app from this messy closet pic

Name: Pack From Closet Tagline: Turn that messy closet pic into a trip-ready packing list. Flow: 1. Upload your closet photo and enter destination, days, and weather. 2. Review auto-detected tops,…Show allShow less

Name: Pack From Closet

Tagline: Turn that messy closet pic into a trip-ready packing list.

Flow:

  1. Upload your closet photo and enter destination, days, and weather.
  2. Review auto-detected tops, bottoms, shoes, and layers.
  3. Check off items to get your final packing checklist.

Models:

  • SAM 3.1 for finding, outlining and counting clothes in the photo
  • Muse Spark 1.3 for image understanding and structured packing answers
  • Muse Wonder 1.0 for long multi-step trip planning

Limits:

  • Only sees visible items, not inside drawers or laundry bins
  • Can mix up similar colors or layered clothes in clutter

Cost per run: $0.012

3

I run a small bakery and I need an app that counts cracked eggs on the conveyor belt from this morning photo, I tried tallying by hand and kept losing track around 200.

Name: Egg Crack Counter Tagline: Stop hand-tallying, get your cracked count in seconds. Flow: 1. Upload this morning's conveyor belt photo. 2. Mark one cracked and one good egg to guide detection. 3…Show allShow less

Name: Egg Crack Counter
Tagline: Stop hand-tallying, get your cracked count in seconds.

Flow:

  1. Upload this morning's conveyor belt photo.
  2. Mark one cracked and one good egg to guide detection.
  3. Review the outlined count and save the tally.

Models:

  • SAM 3.1 for finding, outlining and counting objects in photos
  • Muse Spark 1.3 for text and image understanding and structured answers

Limits:

  • Overlapping or stacked eggs around 200+ can be undercounted.
  • Blurry photos or glare on wet shells lowers accuracy.
  • Won't distinguish hairline cracks vs dirt without clear close-up.

Cost per run: $0.008

4

Photo of my garden bed attached, I want an app that outlines tomato seedlings vs weeds so I do not pull the wrong thing, my neighbor tried to help and we both guessed wrong last spring.

**Name:** Seedling Saver **Tagline:** Keep the tomatoes, pull the weeds. **Flow:** 1. Snap a photo of your garden bed. 2. See tomato seedlings outlined in green vs weeds in red. 3. Tap any outline…Show allShow less

Name: Seedling Saver

Tagline: Keep the tomatoes, pull the weeds.

Flow:

  1. Snap a photo of your garden bed.
  2. See tomato seedlings outlined in green vs weeds in red.
  3. Tap any outline for pull-or-keep advice.

Models:

  • SAM 3.1 for finding, outlining and counting objects in photos
  • Muse Spark 1.3 for text and image understanding and structured answers

Limits:

  • Young seedlings and weeds can look alike, mistakes happen.
  • Poor light, overlap, or blurry photos reduce accuracy.
  • Does not identify disease or nutrient issues.

Cost per run: ~$0.008

5

Need an app to remove the windshield glare from this car damage photo for my insurance claim, the adjuster rejected the first one and I only have my phone pics.

**Name:** Glare-Free Claim **Tagline:** Clean up phone glare to show damage clearly. **Flow:** 1. Upload your car damage phone pic and mark the glare area. 2. Preview the de-glared version and comp…Show allShow less

Name: Glare-Free Claim

Tagline: Clean up phone glare to show damage clearly.

Flow:

  1. Upload your car damage phone pic and mark the glare area.
  2. Preview the de-glared version and compare side-by-side with original.
  3. Download a pack with original + enhanced copy and disclosure note for adjuster.

Models:

  • Muse Spark 1.3 for text and image understanding and structured answers to detect glare vs. damage
  • Muse Image 1.0 for generating or editing images to reduce glare

Limits:

  • Cannot recreate damage hidden under glare, only reduces reflection.
  • Edited photos may still be rejected - insurers often require originals.
  • Works only on phone-quality stills, not heavy blur or video.

Cost per run: ~$0.015

Showing the first 5 of 40.