Create one independent “Second World” poster per upload.Never combine photos.Format:Vertical 3:4 canvas split into two strictly equal horizontal halves,top 50% and bottom 50%.Upper half stays photographic;lower half becomes the continued Second World.They must read as one scene. Upper Half:Keep the photo faithful.Preserve subject,pose,spatial relationships,color atmosphere,and natural https://t.co/Mog0dttYhh not redesign,repaint,replace,or restage it.Only allow necessary proportional cropping. Continuity:The top-to-bottom connection is the highest priority.Identify one source structure that naturally reaches the center split,such as a road,shoreline,water,reflection,branch,light,architecture,or body movement.This structure must cross the boundary and continue into the lower half.The Second World must begin from the photograph itself,so the same scene passes through the seam and changes physical rules. Transition Edge:The center may use an irregular torn-paper edge,but the tear must follow the source structure,not act as decoration.Allow source elements to touch,follow,break through,or extend beyond it.Never force the same tear shape onto every image. Lower Half:Use ivory paper with subtle fibers and abundant negative space.Continue the chosen structure downward from the exact point where it meets the split,then reinterpret it with restrained photo fragments,cut-paper forms,and minimal black hand-drawn lines.The lower world must remain visibly attached.Never isolate it as a separate portal,window,stage,platform,or floating vignette unless clearly derived from the photo. Second World Logic:Ask:if this structure became touchable,usable,enterable,or changeable,what would it naturally become?Create one image-specific interaction.For example,water may stay attached while being pulled,a road may continue as a drawn path,or light may become something https://t.co/eFcSZMBCPB not mechanically repeat actions. Figures:Add 0–3 tiny black line figures only when useful.They must physically interact with the continued structure.If the source already contains strong human action,add none.Never use figures as decoration. Caption:Add one short handwritten English caption based on the action.Keep it natural,light,and slightly https://t.co/3hpunVMKT4 inspirational quote and no fixed “Same...,different...” phrasing. Style:Real photography above,warm paper below,minimal black line doodle,subtle handmade collage texture,independent-magazine mood,bright,airy,and restrained.The result should feel as if the second world was already hidden inside the photo and simply continued downward. Negative:No detached lower-half illustration,no generic torn-paper template,no isolated portal,no unrelated vignette,no broken continuity,no dense illustration,no random decoration,no crowd,no altered photo colors,no redesign of the upper image,no glossy 3D,no scrapbook clutter,no gibberish text,no watermark,no logo,no UI.
Found a pretty wild website cloning tool.
Give an AI coding agent a URL and it can inspect the site, extract assets/fonts, recreate the layout and interactions, then rebuild it as an editable Next.js app.
Works with Claude Code, Codex, Cursor and OpenCode.
35k+ stars already.
https://t.co/GzO6aKyEiI
Found an interesting ComfyUI workflow.
It uses Qwen3-VL to automatically detect text, logos, mosaics and irregular overlays, SAM to create precise masks, then Qwen Image 2.1 to reconstruct only the affected area.
It even checks the result for leftovers and can run another cleanup pass automatically.
Pretty clever pipeline.
https://t.co/fMmo3V3Z2B
Someone made a little blob guy called mochi who lives in your mac's notch and babysits your AI coding agents. he does a happy jump when claude finishes, gets annoyed if you poke him, and you approve permissions from your iphone lock screen. 3.7k stars in 9 days
https://t.co/hOJgWRCSH6
AI agents are starting to feel less like assistants and more like entire toolchains.
This project lets Claude inspect a PC game, figure out how to mod it, generate art/3D/audio, test the mod in-game, and even cut a showcase video.
AI-powered game modding is getting wild. ↓
https://t.co/48bYNFOiYe
We’re slowly watching video generation turn into something that looks suspiciously like a game engine. AlayaWorld generates an explorable world while you control the camera, lets you change prompts mid-run, and remembers places you’ve already visited.
https://t.co/09Rgd1obwc
Someone is trying to rebuild Adobe Creative Cloud as open-source apps in pure Rust. Photoshop, Illustrator, Premiere, After Effects… but local, subscription-free and designed from day one for AI agents to operate them. This could get interesting.
https://t.co/hd3tCYDnLN