Corner Shop Chat

Status: Live.

1. Your muse

   Rebuild ofBuilding a multimodal WhatsApp chatbot with Llama 4 (opens in a new tab)Partial

   Make a customer chat for a small shop that understands photos and voice notes. This rebuilds Meta's video "Building a multimodal WhatsApp chatbot with Llama 4" (https://dev.meta.ai/llama/resources/videos/build-multi-modal-whatsapp-chatbot-with-llama-4) on current Muse models. It demonstrates: A business chatbot on WhatsApp that handles text and images. Plan: A web chat for a small business that understands photos and voice notes and can reply with images. Connecting it to WhatsApp needs a WhatsApp Business account, so the rebuild shows the web version and documents the webhook.

2. Show thinking

3. Checks passed

4. 

   The plan: Corner Shop Chat

   A web chat for small shops that answers text, photos and voice notes.

   Can build part of it

   1. Owner adds shop name, products, hours and policies
   2. Customer sends a text question and optionally adds a photo or voice note
   3. App transcribes voice and asks Spark for a grounded answer

   Won't do: Rebuild fit: partial

   See the full plan

   Who it's for

   Small shop owners and their customers who need quick product help.

   How it works

   1. Owner adds shop name, products, hours and policies
   2. Customer sends a text question and optionally adds a photo or voice note
   3. App transcribes voice and asks Spark for a grounded answer
   4. Customer sees the answer plus an optional generated image
   5. Owner copies the webhook guide for later WhatsApp setup

   Screens

   • Shop Setup. Owner enters shop info, products, FAQs and when to generate images.
   • Customer Chat. Customer sends text, photo or voice note and gets answers and images.
   • Webhook Guide. Shows steps and sample payload to connect the flow to WhatsApp later.

   What goes in

   • Text: Shop name, hours, products and policies typed by the owner. Customer questions typed in chat.
   • Image (optional): Customer photo such as a damaged item or product on shelf. JPG or PNG under 4.5 MB.
   • Audio (optional): Short voice note up to 2 minutes. Browser converts it to mono 16-bit WAV before transcription.

   What comes out

   • Text: Helpful answer with shop details and next steps for the customer.
   • Image: One optional reply image such as a product view or simple care visual.

   Models

   Model	Used for	Why
   Muse Spark 1.3	Answer shop questions from text and photos and decide when a reply image helps.	Default reasoning and vision model that reads photos and shop info and returns grounded answers.
   Voice Transcribe	Turn customer voice notes into text for the chat.	Speech to text model for recorded audio files.
   Muse Image	Create simple reply images such as product views or care guides.	Image model that generates or edits one reply image per run.

   Limits

   • Rebuild fit: partial. The web version handles text, photos and voice notes but does not connect to WhatsApp because products cannot use external messaging accounts or keys. It shows a webhook setup guide instead.
   • Shop data stays in the browser localStorage. No accounts or shared inbox.
   • Voice transcripts are plain text with no speaker labels and no word timestamps.
   • Image replies are new illustrations, not exact product photos. Size setting controls aspect ratio only.
   • WhatsApp connection, payments, order tracking and live human handoff are not included.

   From your inspo

   Rebuild of Building a multimodal WhatsApp chatbot with Llama 4. Keeps a small shop web chat that understands photos and voice notes and can reply with images.

   Takes

   • Small business chatbot job that answers text, photos and voice notes
   • Simple customer chat flow with text answers plus optional reply images
   • Practical shopkeeper setup with webhook guide for later WhatsApp use

   Leaves out

   • Video footage and script from the original developer video
   • Meta and Llama names and logos
   • WhatsApp brand and interface design

   Cost per run: about $0.02 (est.). Spark chat about 2k input tokens at $1.25 per million is $0.0025 plus 400 output tokens at $4.25 per million is $0.0017. Voice note of 30 seconds at $0.18 per hour is $0.0015. One reply image is $0.01. Total about $0.0157, rounded to $0.016.

6. Done. Corner Shop Chat is built and verified.

   It's live.

   musemuse-corner-shop-chat-2.vercel.app (opens in a new tab)

   Product pageNew muse

   11 m 38 s · Input 4.9M · Output 50.7k · $3.11 est.