How to Build an AI Pitch Deck in One Session with Stensyl.

Start with the Brief, Not the Slides

The most common mistake in any AI-assisted pitch deck is opening a blank slide template first. Structure comes from story, and story comes from brief. Skip that sequence and you get visually polished slides with nothing coherent to say.

Before a single visual is generated, Stensyl's Research (/research) surface gives you a Perplexity-backed research layer to ground the deck in real context. Whether you're building a pitch for a consumer product launch, an exhibition stand concept for a heritage brand, a brand identity proposal for a retail client, or a game studio seeking publisher funding, the inputs are different but the discipline is identical: understand the market, the audience, and the competitive landscape before you write a word of copy. Independent pitch-deck guides consistently flag this step as the one founders and agencies most often skip, and the one investors most quickly notice is missing.

Once you have that context, open Ray (/ray). Ray is Stensyl's creative-decision assistant. Describe the pitch: its purpose, its audience, the decision-maker in the room, and which surfaces are likely to carry the visual load. For a motion design studio pitching a broadcast package, Ray will point you toward Film and Motion for supporting assets. For a UX team presenting a north-star product vision, Generate and Graphics are the natural pairing. Ray removes the guesswork about where to spend your credits before the session starts.

Then paste the raw brief into Write (/write). This is where the deck's architecture gets built. Write gives you access to six writing models on every plan. For a structured, section-by-section slide outline, Claude Sonnet 4.6 is the natural starting point: it handles business narrative well and produces clean hierarchical outputs. If the deck needs punchy, direct executive framing (a marketing agency presenting a campaign ROI case, for example), switch to GPT-5.5 for a harder register. Both models are available on every Stensyl plan, so the choice is creative, not financial.

Define the deck's core argument in one sentence inside Write before you expand into slides. If you cannot summarise the pitch in a single sentence, the slides will be unfocused regardless of how strong the visuals are. This is not a formatting exercise: it is a thinking exercise, and it is the one that separates decks that get a second meeting from decks that don't.

Independent guidance across pitch-deck workflows is consistent here: a canonical structure (problem, solution, market, traction, business model, team, ask) exists because investors and senior clients have internalised it. Deviating too far increases cognitive load and signals unclear thinking. Use Write to draft a slide-by-slide outline that follows this logic, annotated with the key visual direction for each slide, before you open any generation surface.

Write the Slide Copy Across All Six Models

A pitch deck's copy is not filler between images. It is the argument. The visuals reinforce it. Getting the words right before touching Generate or Film is both a creative discipline and a practical one: long-context LLM calls and video generation are materially more expensive per asset than short text passes, so the sequence matters for your credit budget too.

Write (/write) gives access to all six models from a single picker: GPT-5.4 mini, Gemini Flash, Gemini Pro, GPT-5.5, Claude Sonnet 4.6, and Claude Opus 4.7. Each has a different centre of gravity, and using them in combination is a genuine workflow advantage rather than a novelty.

Model	Best use in a pitch deck session
GPT-5.4 mini	Fast headline rewrites, quick bullet-point alternatives, low-cost iteration
Gemini Flash	Rapid variant generation for subheadings and alternative framings
Gemini Pro	Data-dense market sizing slides, retrieval-heavy analysis sections
GPT-5.5	Punchy executive framing, conclusion-style headlines ("A £12B market growing 24% YoY")
Claude Sonnet 4.6	Structured narrative slides, problem and solution sections, balanced register
Claude Opus 4.7	Tone-checking investor copy, executive summaries, "About the team" slides

A practical pattern: use Gemini Pro to draft the market sizing slide, where data framing and analytical rigour matter most, then switch to Claude Sonnet 4.6 for the problem and solution slides, where the narrative arc needs to carry the reader forward. Test the same headline in two models and pick the version that reads like the client's own language rather than a pitch template.

Draft each slide as a structured block inside Write: headline, one supporting sentence, and the key visual direction. This format does two things. It keeps copy tight, because the constraint forces clarity. And it gives the generation surfaces something precise to work from later: a prompt anchored to a specific headline produces a visual that reinforces the argument, not one that merely looks good alongside it.

For client-facing decks where tone is especially sensitive (a branding agency presenting a full identity system to a legacy retail client, or an interior designer pitching a hospitality refurbishment to a conservative ownership group), run the executive summary slide through Claude Opus 4.7. Opus is Anthropic's highest-capability model and consistently produces copy that mirrors a business register more precisely than mid-tier models. It is available on every Stensyl plan. Use it for the slides where a single word choice affects the room.

Before moving into generation: export the full outline as a structured document. Switching back to rewrite copy mid-image-generation session wastes credits and breaks the flow of the session. The sequence is not arbitrary. Brief, then copy, then visuals. Every deviation from this order has a cost.

Generate the Visuals Slide by Slide

With the copy locked and the outline exported, every visual decision already has a brief. This is what makes the generation phase fast: you are not exploring, you are executing against defined slide requirements.

Generate (/generate) handles individual hero images. The scope here is genuinely wide across disciplines. A consumer goods team needs a product render showing the hero SKU in its retail context. An exhibition designer needs a spatial visualisation showing how a brand installation reads in a large venue. An automotive studio needs a vehicle colourway study across three exterior options. A game studio needs key art frames that establish tone, genre, and world. Generate handles all of these from a single surface, and the model doing the generation work is consistent across your plan tier.

Keep image prompts anchored to the slide copy already written. A prompt that references the slide's headline directly produces a visual that reinforces the argument. A prompt that simply describes "a futuristic product on a white background" produces a visual that decorates around it. The difference matters in the room.

Graphics (/graphics) covers the designed elements that sit around those hero images: layout backgrounds, icon sets, typographic treatments, and the slide furniture that gives a deck its visual language. Where Generate produces the centrepiece, Graphics handles the system. A deck where both surfaces are used in combination tends to read as cohesive rather than assembled.

Before committing to a visual direction, use Moodboards (/moodboards). Assembling reference for the deck's colour palette, material language, and tone before running Generate prompts keeps the visual output consistent from slide to slide. A product pitch that drifts between warm analogue photography and cold digital renders loses visual authority, even if each individual image is strong. Locking the vocabulary early via Moodboards prevents that drift. It also gives you a reference document to share with a client or stakeholder before generation begins.

For pitches that include motion (a motion design studio presenting a broadcast package, a UX team embedding an interactive prototype walkthrough, or a film production company pitching a title sequence concept), Film (/film) and Motion (/motion) produce the supporting assets. Film handles multi-scene cinematic sequences. Motion, built on Remotion, handles programmatic motion graphics with clean export. For live pitches, short muted loops embedded in the deck tend to work more reliably in the room than longer narratives: they demonstrate the capability without creating a technical dependency on stable playback. Note that AI video generation is significantly more expensive per asset than still images across all platforms, which is another reason to lock story and copy first and generate visuals only to the specific brief.

One practical note on limitations: brand-accurate logos, complex UI screens, and real-world product likenesses at precise tolerances will typically need refinement after generation. For technical fields, AI images get you alignment and speed in the room. Credibility on the specifics still depends on human review and, where relevant, real geometry or real data.

Use Canvas to Chain Copy and Visuals Together

The four surfaces used so far (Research, Write, Generate, Graphics) are powerful individually. Canvas (/canvas) is where they become a single connected workflow rather than four separate sessions.

Canvas is Stensyl's node-based workflow editor. Pipe the Write output into an LLM Chat node to refine individual slide headlines, then route the refined copy into Image Generate nodes to produce slide-specific visuals in one connected chain. The visual logic of the deck becomes auditable. You can see exactly which copy drove which image, and re-run a single node without rebuilding the entire session.

The LLM Chat node in Canvas supports the same six writing models as Write. You can switch model mid-chain without leaving the editor. This is useful when one section of the deck needs a different register from the rest: the market slide might need Gemini Pro's analytical framing while the team slide needs Claude Opus 4.7's warmer, more human tone. Change the model at the node level and the chain re-runs from that point.

For a multi-section deck, build a separate node cluster per section. Problem, solution, market, team, ask: each cluster contains its own LLM Chat node and its own image generation nodes. This structure mirrors how experienced founders and agencies already work: they iterate on problem and solution framing independently from market sizing, and they revise the ask separately from the team narrative. Canvas makes that natural separation explicit rather than forcing a linear process.

Node-based creative workflows are not new. VFX artists have worked in node graphs in Nuke and Fusion for decades. Motion designers using ComfyUI pipelines for concept art are already accustomed to chaining prompts, operations, and outputs in a graph structure. Canvas applies the same logic to a pitch-deck session: each node has a defined input and a defined output, and the chain is readable at a glance.

For decks that will be iterated across multiple client rounds (a brand consultancy presenting to a client who revises the brief after each stakeholder review, or an architectural practice responding to planning feedback before a second committee presentation), save the Canvas chain after the first session. When the brief changes, re-run the relevant cluster rather than rebuilding prompts from scratch. This is the practical advantage of a connected workflow over the typical pattern of copy in one tool, images in a second, assembly in a third: the chain remembers the logic.

Assemble and Publish Without Leaving the Platform

The deck is built. The copy is refined, the visuals are generated, and the Canvas chain has connected them. The final phase is assembly, review, and getting the work in front of the right people.

Editing (/editing) is Stensyl's frame-level image editing studio, available on desktop. Use it for final polish: adjusting crops on a hero visual, refining contrast on a generated product render, or cleaning up an asset before it goes in front of a client. This is not a wholesale redesign step. It is the quality check that separates a deck that feels finished from one that feels assembled.

Projects (/projects) keeps all deck assets, brand references, generated images, and team notes in one shared workspace. A second designer picking up the project mid-session does not need to hunt across folders or request access to five separate tools. The brief, the outline, the generated visuals, and the Canvas chain are all in the same place. For studio teams working across disciplines (a product design studio where one person handles copy and another handles visualisation), this shared structure is the difference between collaboration and coordination overhead.

For remote stakeholders who will not be in the room, Web (/web) lets you publish the pitch as a microsite. Building the site is available on any Stensyl plan. Going live with a custom domain requires Pro or Studio. For a marketing team sharing a campaign pitch with a global client, or a UX agency sending a north-star product vision to a distributed executive team, a published microsite is a more controlled environment than an emailed PDF: the assets render at full quality, the sequence is preserved, and you control the viewing experience.

Social (/social-studio) produces formatted versions of key slides as standalone social assets. A graphic designer sharing process work from a completed identity system, a marketing team announcing a campaign after the pitch lands, or a product studio teasing a launch with key art frames: Social handles the format translation from deck asset to platform-ready post. The visual language built in Generate and Graphics carries through without requiring a separate production pass.

Session discipline is not just a creative preference. It is a credit discipline. Brief and copy are inexpensive. Image generation costs more. Video generation costs significantly more. Running the session in the correct sequence (Research and Write first, Generate and Graphics second, Canvas to chain, Editing to polish, Web or Social to publish) is faster and cheaper than the alternative on almost every modern AI platform, and Stensyl is no exception.

The case for a one-platform approach is not theoretical. The typical "copy in Notion, images in Midjourney, slides in Keynote, video in Runway" workflow involves five context switches, five credit or subscription systems, and five separate export and import steps every time the brief changes. When a stakeholder requests a directional revision after the first presentation (and they will), rebuilding across five tools is a half-day job. Rebuilding a Canvas chain and re-running the affected nodes is not.

A pitch deck built across one session in Stensyl, following the sequence set out here, is not only faster to produce. It is faster to iterate, easier to hand off, and simpler to publish. The session discipline is the workflow. Run the brief through Research and Write. Test the copy across the model picker. Generate visuals slide by slide. Chain everything in Canvas. Polish in Editing. Publish in Web or Social. That is the session. One platform, one credit system, one place to pick up when the brief changes again.