Combining Modalities: Multimodal Workflows

Update November 2, 2026 — Le Chat becomes Vibe. Since May 28, 2026, Le Chat has been rebranded as Vibe (still at chat.mistral.ai): a single agent for work and code, carrying over your conversations, settings, and subscriptions. The legacy Le Chat experience remains accessible in Chat mode; in Work mode, agents have become skills (invoked with @), memories serve as the knowledge base, and deep research is now the Deep Research skill. The plans are now called Mistral Free, Pro, Team, and Enterprise. This lesson uses the name Vibe; note that some menus described here may have changed names or locations.

The Power of Combined Multimodality

Throughout this course, you have explored three core capabilities of Mistral AI: vision for image analysis, audio for transcription and speech synthesis, and document processing with intelligent OCR. Individually, each is impressive. But combining them in a single workflow unlocks their full potential: sources intersect and complement each other, producing results that exceed what any single modality could achieve alone. This lesson shows you how to orchestrate these combinations.

Image + Text: Beyond Description

The most common use case involves enriching an image with precise text instructions. The difference between a weak prompt and a rich one is dramatic. Take a screenshot of a PowerPoint slide: “Describe this image” yields a bland paraphrase. But “Here is a slide from our investor presentation. Rephrase the main message for a general audience” or “This slide shows our quarterly results. Identify the top 3 figures and suggest a catchy title” leverages both the model’s visual understanding and its writing ability—making the output immediately usable.

The same principle applies to field photos. “Here is a photo of our booth at the trade show. Suggest improvements to attract more visitors” turns the model into a consultant; “This photo shows our current office layout. Propose a reorganization to improve collaboration” makes it a space planner. In both cases, the text context directs the model’s focus.

You can also combine multiple images with a comparative instruction: three design mockups accompanied by “Rank these mockups by visual clarity and justify your ranking,” or a before-and-after renovation photo with “Estimate the percentage of progress and identify remaining tasks.”

Audio + Document: The Smart Summary

A particularly powerful workflow crosses audio transcription with document analysis. The classic scenario is a meeting: first upload the agenda (PDF) to Vibe, then the meeting recording, and finally ask: “Here is the agenda and the recording of our meeting. For each agenda item, summarize what was decided and list the action items.” The model cross-references both sources to produce a structured, point-by-point summary—whereas transcription alone would give you a long verbatim transcript to edit, and the agenda alone would be an empty shell.

The student variant works just as well: upload the course PDF slides, then the professor’s audio recording, and ask “Create comprehensive revision notes by combining the slides with the professor’s oral explanations.” The slides provide structure and exact definitions; the audio adds explanations, examples, and emphasis. The final notes are far richer than what either source could provide separately.

Four Professional Workflows to Replicate

For competitive intelligence, take screenshots of three competitors’ homepages, upload them to Vibe, and ask: “Compare these 3 homepages. Identify the strengths and weaknesses of each in terms of messaging, design, and call to action.”

For client onboarding, upload the signed contract (PDF) followed by the client brief (document or image), and ask: “Summarize the contractual commitments and cross-reference them with the brief’s expectations. Are there any inconsistencies?” Catching these mismatches early prevents discovering mid-project that a sales promise isn’t in the contract.

For event analysis, combine three sources: event photos, audio recordings of speeches, and the program (PDF), then ask: “Create a post-event report highlighting key moments, memorable quotes, and an evaluation of how well the program was covered.”

For internal training, reverse the flow: record your oral explanation of a process, let Vibe transcribe it, ask “Turn this transcription into a structured procedure guide with numbered steps,” then add screenshots with “Integrate descriptions of these screenshots as illustrations for each step.” In four messages, an improvised explanation becomes formal documentation.

Running the Machine Smoothly

A few habits make these workflows much smoother. Contextualize every file you upload—“This is the agenda,” “Here is the corresponding recording”—so the model associates sources correctly. For complex workflows, proceed step by step: rather than sending everything in one message, build the conversation message by message, validating each step before moving to the next. Name your files explicitly before uploading: “report-q3-2026.pdf” informs the model; “document.pdf” does not. And remember that voice mode is itself an input modality: for long instructions, dictate them rather than typing a paragraph.

Key Takeaways

• Combining image + text + audio + document in a single workflow multiplies possibilities
• The most powerful workflows cross-reference multiple sources: meeting audio + agenda, photos + client brief
• Contextualize every uploaded file to help the model make the right connections
• Proceed step by step in complex workflows rather than sending everything at once
• Voice mode can serve as an input tool for detailed instructions