I got tired of stock photos that look like stock photos. So I built <Inktag/>

drop-in img replacement is the part that'd get me to try this. most AI image tools make you leave the component tree entirely - this stays in JSX. do you cache renders when the same prompt fires twice?

Interesting positioning — the “same JSX surface as <img/>” immediately made the idea click for me. The bigger insight around consistency/brand drift feels more compelling than “AI image generation.”

One thing I’m curious about: how are you thinking about the moment when users need something very specific or campaign-driven that intentionally breaks brand consistency?

Feels like there’s a tension between “locked brand identity” and creative flexibility. Is the main ICP content-heavy sites/newsletters where consistency matters more than originality, or do you see agencies/marketing teams adopting this too?

The 100-seat private beta with hand-selected use cases is the part of this I'm watching most closely.

Most launches optimize for top-of-funnel volume — get as many free signups as possible, then panic about conversion later. You're doing the opposite. By gating on "what will you generate," you're filtering for people who already know their visual identity and have a real workflow gap. The feedback you get from those 100 will be worth more than 10,000 random signups.

The pricing question I'd want answered: per-image generation cost on Gemini isn't zero, and your caching layer (<2s for repeat prompts) only helps after the first hit. How are you thinking about the unit economics when a content site generates 50 unique hero images a month vs. 500? Is the pricing per-image, per-seat, or tied to brand-config count?

Also curious whether the "brand.config" concept extends beyond images eventually. The same locked-variable pattern could work for video thumbnails, social card generators, even product mockups. The infrastructure framing is the real moat here.

Applied with one of my own sites. Rooting for this.

Love this approach a lot. Treating it like “brand infrastructure” instead of yet another image tool really clicks, especially the locked palette/style + JSX drop‑in. As someone who’s spent way too long fighting stock photos and prompt drift for blogs, this feels like the right abstraction layer. 

This is a really interesting approach especially locking brand rules (palette, style, exclusions) at the config level instead of relying on prompt consistency.

The JSX-style integration also feels very natural for React workflows, which could make adoption easier for dev teams building content-heavy sites.

Curious how you’re handling edge cases where prompts conflict with brand rules, and whether users can override style constraints when needed.

Overall, this feels closer to “design system for generated images” than just another AI image tool

This is actually a smart angle. Most AI image tools focus on generating “better images,” but the real pain for content sites is consistency across dozens or hundreds of posts. Locking palette/style at the config layer instead of relying on prompt engineering makes a lot of sense.

The <Inktag prompt="..." /> approach is also clean from a developer perspective — feels more like infrastructure than another design tool. I can see this being useful for blogs, SaaS docs, newsletters, and niche media sites that want recognizable visuals without constantly touching Photoshop or Figma.

Curious how you’re handling long-term brand drift though. If the underlying model changes over time, do older prompts still regenerate with nearly identical style consistency, or are you versioning generations/models per project?

Looks interesting to me!
Can this help product based D2C entrepreneurs for their authenticity based content?

Love the approach of locking brand variables at the config layer instead of fighting prompts every time. The <Inktag prompt="..."/> DX is super clean — same mental model as <img> but no more stock-photo fatigue.

Curious how you’re handling model routing under the hood. Are you dynamically selecting the best model per prompt, or does the user pin one? Also, how do you prevent subtle brand drift over months when models get updated — do you version the brand config somehow?

Applied with our newsletter’s visual identity. This could genuinely save us a couple of hours every week. Congrats on the beta, Gautam!

Locking the variables that shouldn't change at the config layer instead of the prompt layer is the right architectural call. Everyone else is trying to solve brand consistency by writing better prompts, which is basically hoping for the best. The JSX surface being identical to img is also a smart DX choice zero mental overhead to drop it in. Applying for the beta.

Really interesting product. Turning something like this into a product-led business is not easy, especially with hardware + distribution involved. Curious how you’re handling user acquisition and whether the demand is mostly dev / maker community or broader consumers. Either way, cool to see it being built in public 👏

I like this...consistency over time is such a struggle with AI images.