Search documentation
Find guides and API references.
Image generation
I use this page to show you my full CanvasFlow image system, from first setup to conversion-ready delivery inside Obsidian.
What you get from my image workflow
I built this process for operators who need visuals that convert, not random one-off renders. If you follow the runbooks below, you get repeatable creative output, faster approvals, and cleaner campaign execution.
What you get: You get one approved hero image plus three channel variants (site, social, and docs) from one prompt stack.
Why I care: I remove launch blockers caused by missing creative, so your publish window stays intact.
What you get: You get a structured variant loop that cuts subjective back-and-forth during review.
Why I care: I keep stakeholders aligned on objective checks (composition, readability, CTA space) instead of opinion loops.
What you get: You get prompt + seed metadata archived beside every final image for deterministic reruns.
Why I care: I turn one successful campaign visual into a reusable production template you can scale.
Setup I require before first render
I never start prompt work before this setup is complete. It prevents silent failures and makes every winning render reproducible.
- •I enable SystemSculpt Canvas enhancements (experimental) under Settings → SystemSculpt (desktop only).
- •I open Image Generation and run Test image generation API to sync the live provider catalog.
- •I choose a default image model with the search button beside Default image model so every CanvasFlow run starts runnable.
- •I lock an output folder and naming pattern (for example
YYYY-MM-DD-campaign-v1) before the first production batch. - •I keep Job poll interval (ms) and Write metadata sidecar enabled so retries and reproducibility remain stable.
End-to-end runbooks I use in production
Each runbook covers the full path from brief to published asset. Pick one based on what you need to ship, then execute it exactly.
What you get: You finish with a hero image, one backup variant, and prompt metadata ready for handoff.
Conversion outcome: I design for copy-safe negative space so headlines and CTA buttons stay readable above the visual.
- I define the brief: audience, conversion action, emotional tone, and must-keep brand constraints.
- I generate 6 low-cost drafts at the target aspect ratio and reject anything without clear focal hierarchy.
- I refine the best 2 prompts, then run delivery quality on the winner with identical framing constraints.
- I export image + metadata sidecar to the launch folder and embed it in the page draft for QA.
- I run desktop/mobile visual QA before I mark the asset as publish-ready.
What you get: You finish with a consistent multi-image sequence that mirrors the operational flow in the guide.
Conversion outcome: I reduce confusion in onboarding docs, which lowers drop-off before users reach paid-value actions.
- I split the workflow into 3 to 5 teachable frames so each visual carries one clear instructional job.
- I generate each frame with the same style anchors so the full sequence feels intentionally unified.
- I reject frames that hide UI structure or overload text inside the image itself.
- I export the sequence into a docs-specific folder and attach usage notes for future updates.
- I verify every embed renders clearly in both light and dark Obsidian themes.
What you get: You finish with a channel-ready batch that keeps one message theme while adapting format and composition.
Conversion outcome: I preserve message consistency across channels, which improves recognition and click intent.
- I lock one campaign message and define channel-specific format constraints before prompt writing.
- I generate small variant sets per channel instead of one giant mixed batch.
- I score each variant against a simple gate: clarity, relevance, brand alignment, and CTA compatibility.
- I export winners into a dated campaign folder and archive rejects for auditability.
- I log top-performing prompts so next week starts from validated creative patterns.
Prompt systems I reuse for predictable quality
I avoid vague one-liners. Structured prompts give me predictable composition, faster approvals, and more stable outcomes across campaigns.
| Pattern | Best for | What you get | Prompt template |
|---|---|---|---|
| Conversion hero frame | Landing pages, pricing sections, launch banners | A high-clarity hero with intentional copy space and consistent visual hierarchy. | |
| Workflow explainer strip | Docs walkthroughs, onboarding guides, process pages | A multi-stage visual sequence that explains the process without extra copy bloat. | |
| UI concept scene | Feature previews, roadmap announcements, product updates | A realistic UI-style visual you can use before final engineering polish exists. | |
| Controlled style variant | Campaign testing, creative direction reviews, mood variants | Multiple mood options while preserving subject identity and framing. | |
Quality gate I run before publish
I never publish directly from a fresh render. This gate keeps visual quality and conversion intent consistent across every channel.
- •I confirm the focal subject is readable at thumbnail size.
- •I verify there is clean copy-safe space where headlines or CTA blocks need to sit.
- •I reject outputs with malformed details, over-smoothed textures, or accidental text artifacts.
- •I keep only one approved final and one backup variant per placement to prevent asset sprawl.
- •I store the winning prompt, seed, model, and ratio in the sidecar metadata note.
If this gate fails, I rerun one controlled prompt change instead of rewriting the full brief.
Troubleshooting path I follow before escalation
Most image failures are recoverable in one cycle when I adjust one variable at a time. I escalate only after I gather clean reproduction evidence.
I include the exact prompt, model, aspect ratio, and quality preset.
I attach the failed output plus timestamp so support can replay the run path.
I run the Troubleshooting guide first, then I attach that result summary with my escalation.
Weekly image ops cadence I use
This weekly rhythm keeps my image library clean, reusable, and aligned to conversion goals over time.
- •I run a weekly prompt audit and promote only high-performing prompts into my reusable template list.
- •I delete failed drafts and duplicates so search results stay clean and teams do not reuse weak assets.
- •I annotate each published image with channel, owner, and publish date for fast handoffs.
- •I capture one lesson per campaign so my next run starts from evidence instead of guesswork.
1. I review the week’s top-performing visuals and archive reusable prompt patterns.
2. I remove low-signal variants to keep future selections fast and clean.
3. I sync final image links into notes and campaign planning docs.
4. I align image plans with my chat workspace and next content priorities.
