Ohans Emmanuel

@hewarsaber And make it move :)

@hewarsaber Tried a prompt here, but still pretty cool: bluebag.ai/skillet/F30SbQ…

Prompt: { "prompt": "A hyper-realistic {{object}} rendered as a sculptural object carved entirely from solid, optical-grade crystal glass. The object preserves its true proportions and recognizable silhouette, with precise geometry and clean edge definition appropriate to its form. It is floating in space at a slight dynamic angle, not perfectly straight, allowing light interaction across multiple surfaces. A single focused white light source enters the object from one direction, refracting through the internal volume and producing layered internal reflections, chromatic aberration, and physically accurate spectral dispersion along edges, corners, and internal transitions. Light bends, splits, and bounces naturally within the material, creating complex but controlled optical behavior. Inside the glass, subtle volumetric depth is visible, including faint internal reflections, refractive distortions, micro-scratches, and tiny dust particles suspended within the material. A small, intense specular highlight appears on a sharp edge or curvature where the light strikes most directly, producing a crisp optical spark. Reflections are sharp and physically accurate. The glass feels heavy, dense, and premium — never plastic, never stylized, never neon. The background is a pure black void with soft falloff and no visible environment. Lighting is minimal and studio-controlled, designed to emphasize form, refraction, internal depth, and edge clarity rather than brightness. No text, no logos, no markings.", "style": { "lighting": "Low-key studio lighting with a single directional white light source", "materials": "Optical-grade crystal glass with physically accurate refraction and dispersion", "surface_detail": "Subtle micro-scratches, internal dust, realistic volumetric imperfections", "background": "Pure black void", "mood": "Minimal, futuristic, scientific, museum-grade artifact", "color_behavior": "Rainbow dispersion only from white light refraction, never artificially colored", "Moodboard": "Attached Images" }, "parameters": { "object": "iPod", "orientation": "slightly tilted, dynamic perspective", "refraction_intensity": "high but physically accurate", "edge_sharpness": "crisp relative to object geometry", "spark_intensity": "small, sharp specular highlight", "glass_thickness": "solid, heavy volume", "imperfection_level": "very subtle" }, "medium": "Ultra-realistic 3D render", "size": "1080px by 1080px" }

Different prompt, different model (Flux), one agent Skill Copy prompt here ⬇️ bluebag.ai/skillet/F30SbQ… Use the Skill here: skills.sh/bluebagai/skil…

@favibes1 Yea! Looking to add more video prompts eventually But works great for the supported image models

@OhansEmmanuel Nice! Have you tested consistency across several models with the same superprompt?

i built a Skill for consistently great AI images it's optimised for NanoBanana, Flux, Midjourney, GPT Image, Ideogram, Seedream, and 6+ more models the flow: 1. type a basic idea: "couple at sunset" 2. the Skill generates a structured superprompt with lighting, camera, wardrobe, pose, anatomy 3. use in image model → studio-quality result The gap between what you typed and what you get back is honestly absurd. 100% free & open source.

Spending hours reviewing your teammate's AI-generated code? Review the plan that produced it instead. Code is cheap now. One bad decision in the plan → hundreds of lines of wrong code. Review the input, not just the output.

@wesbos This Agent Skill may be what you need: skills.sh/bluebagai/skil… It's super optimised for nano banana

Who has some tips on getting consistent styles out of Nano Banana? Seems the more I prompt and provide examples, the worse it gets?

@prhaizze 💯

@OhansEmmanuel Saves me the stress of manually inputing prompts that almost never reflect what I have in mind. Will definitely give this a try.

Skill on Github: github.com/BluebagAI/skil… use it. It's free.

see full prompt and interaction in the video here: bluebag.ai/skillet/mW4_Ex…

@Jaan @rtwlz That’s sick! Damn

@rtwlz For reference, epsteinsimages.com is hosted using Workers, images in a R2 bucket, and KV for caching. It has received ~302M pageviews last 30 days. Keep in mind I host a lot of stuff on CF, so this invoice isn't 100% from this site alone.

Jmail has crossed 450M pageviews, and our Vercel bill has exploded, even after tons of cache mitigation. Many kind people chipped in to cover the bill, but it’s not sustainable. What are solid, cheaper alternatives?

Damn. Since most of these are static (no new “emails”), they could mostly be static assets

Not real, AI and video combined Nano Banana Pro and @Kling_ai Both generated from the superimage prompt Skill. Works with videos! 100% open-source dropping soon,

@favibes1 🔥

@OhansEmmanuel Great! Looking forward to this!

Not real, AI gpt-image + super-image prompt Skill Sharing Skill later today 100% open-source and free to use

Not real, AI Nano banana pro and superimage-promt Skill. Releasing Skill tomorrow. Thought I'd try with something more 🌶️ ....

@feelsgood888 I see, thanks. will give it a tr

@OhansEmmanuel haven’t tried it, ease of use is probably a big one (less boilerplate but less flexibility). they have a built-in agent that you can drop in

ngl i’m surprised it took so long for a company like this to pop up, generative ui has been a thing for 2+ yrs now (w/ vercel’s ai sdk)

@unicodeveloper Brilliant name :)

I built an open-source McKinsey for people that don't have $2mil/yr for billable hours💼 Consultants can now collapse hours of research into 5/10 mins... -> Generate deliverables (DOCX/PDF/CSV) -> Market Analysis -> Due Diligence & Competitive Intel 🔗 consultralph.com

@elie2222 @posthog Brilliant software, opinionated UX But it could be much simpler (UX-wise), agree.

@posthog I'm a fan of PostHog, but definitely feel like you guys have added 500 features that I won't touch. Even though they could potentially be useful.

Strange to say, but some would argue Cursor shipped too fast some months ago Seemed they've now slowed down / prioritised reliability