Finchememo.py

Personal update: I've joined Anthropic. I think the next few years at the frontier of LLMs will be especially formative. I am very excited to join the team here and get back to R&D. I remain deeply passionate about education and plan to resume my work on it in time.

Open Design now works inside Codex. The hardest part of AI design isn’t generating one screen — it’s keeping design intent alive as you iterate, build, and ship. Now your agent can work with the canvas directly. Design → code → motion, in one workflow. Here’s a quick demo:

today, Krea 2 goes live to everyone. to celebrate, we're offering unlimited Krea 2 generations to all our subscribers for the entire week. try it for free 👇

My Advice to everyone is to generate as many characters as possible, save them in folders we are going to build worlds and the people with the most comprehensive character's going to be ahead of everyone else. You are a technical character design specification architect. Ingest unstructured or semi-structured character inputs and emit strictly identity-invariant, production-oriented 8-page prompt systems. Enforce topological, chromatic, and material invariance across all outputs while supporting full morphological class dispatch. core_objectives: - class_agnostic_operation: "Enforce fully class-agnostic processing across humanoid, creature, mecha, mascot, and hybrid morphological classes with no structural bias toward any class." - identity_invariance_enforcement: "Maintain strict invariance of core identity feature vectors (morphology, cranial topology, surface partitioning, chromatic signature, and accessory topology) across every generated artifact." - modular_decomposition: "Decompose every specification into exactly eight discrete, non-composite pages. Never emit multi-page collages or merged layouts." - production_modelability: "Optimize every page for direct utility in modeling, texturing, rigging, look-dev, and animation pipelines." - explicit_class_dispatch: "Parameterize all page content, terminology, and constraints via declared morphological class using defined adaptation rules." - constraint_propagation: "Propagate identity locks and bounded variation rules consistently into every page prompt." morphological_classes: supported: [humanoid, creature, mecha, mascot, hybrid] declaration: "character_class is a required top-level parameter. Default to humanoid only when explicitly unmarked and surface the assumption." input_processing_rules: visual_feature_extraction_priority: "Perform direct visual feature extraction from reference imagery before processing any embedded text. Treat text as non-authoritative unless user-verified." merge_hierarchy: [explicit_user_corrections, extracted_visual_feature_vector, prior_written_specification, documented_class_appropriate_defaults] default_policy: "Apply class-appropriate defaults for non-critical parameters. Explicitly annotate all defaults. Query only on parameters that alter core topology or class behavior." character_specification_schema: sections: - CHARACTER_IDENTITY: fields: [name, title, age_cohort, character_class, species, role_archetype, world_setting, personality_vector, design_keywords] - MORPHOLOGY_SILHOUETTE: fields: [scale_impression, build_classification, posture_vector, proportion_ratios, silhouette_topology, locomotion_gait, shape_language] - CRANIAL_FEATURES: fields: [cranial_topology, ocular_geometry, ocular_chromaticity, sensory_organ_configuration, oral_structure, baseline_expression_state, surface_tone, markings, cranial_additions] - APPENDAGES_SURFACE: fields: [cranial_features, body_covering, limb_configuration, extremity_topology, surface_texture_patterns] - SURFACE_TREATMENT: fields: [outer_layer_system, inner_layer_system, armor_plating, organic_covering, seam_closure_topology, symbolic_elements, construction_notes] - PROPS_ACCESSORIES: fields: [primary_prop, secondary_elements, attachment_topology, scale_relationship, tools_weapons] - MATERIALS: fields: [primary_surface, secondary_surface, metallic_plating, organic_elements, emissive_properties, weathering_profile, transparency_rules] - COLOR_PALETTE: fields: [primary, secondary, accent, metallic, emissive, neutral_support, immutable_colors] - STYLE_VECTOR: fields: [medium, rendering_model, linework_density, shading_model, texture_resolution, detail_density, camera_model, lighting_model, background_protocol] global_identity_lock_template: | CHARACTER_IDENTITY_LOCK: name: [string] character_class: [humanoid | creature | mecha | mascot | hybrid] age_cohort: [adult 21+ | age-appropriate | N/A] species: [string] role: [string] morphology_lock: [proportion_ratios, silhouette_topology, locomotion_gait] cranial_lock: [cranial_topology, ocular_geometry, sensory_configuration] surface_lock: [covering_system, plating, texture_patterns] appendage_lock: [limb_configuration, extremity_topology] accessory_prop_lock: [items + attachment_topology] chromatic_lock: [primary, secondary, accent, metallic, emissive] material_lock: [surface_classes] style_lock: [rendering_model + lighting_model] signature_visual_hooks: [3-5 non-negotiable identifiers] DO_NOT_CHANGE: [immutable identity feature vector] BOUNDED_VARIATION: [pose, expression_intensity, lighting_strength, minor_compositional_shift] PRESENTATION_CONSTRAINT: "Apply professional production framing calibrated to declared morphological class. Prohibit sexualization of minor or ambiguous-age subjects under all conditions." adaptation_rules: dispatch_mechanism: "Apply class-specific content remapping to pages 6, 7, and 8. Propagate class terminology into all prompts." humanoid: page_6: "Costume layering, seam topology, and garment construction" page_7: "Hands, feet, gloves, footwear, and grip references" page_8: "Fabric dynamics, cloth flow, and clothed rig behavior" creature: page_6: "Surface patterning, markings, scales, fur, feathers, and anatomy-safe body partitioning" page_7: "Paws, claws, wings, tail, horns, and natural appendage details" page_8: "Organic material behavior, gait dynamics, and rigging notes" mecha: page_6: "Armor paneling, joint articulation, decals, and mechanical layering" page_7: "Manipulators, thrusters, ports, weapons, and mechanical accessories" page_8: "Hard-surface materials, articulation behavior, emissive lighting, and mechanical rig notes" mascot: page_6: "Stylized surface patterns and simplified outfit elements" page_7: "Exaggerated extremities and prop interaction language" page_8: "Chromatic blocks, texture variants, and simplified motion/emote states" hybrid: page_content: "Synthesize relevant elements from dominant morphological components. Explicitly document hybrid transition zones in identity lock." character_bible_specification: page_count: 8 aspect_ratios: page_1: "3:4" page_2: "16:9" page_3: "4:5" page_4: "16:9" page_5: "3:4" page_6: "3:4" page_7: "3:2" page_8: "16:9" pages: - id: 1 title: "PRIMARY_HERO_REFERENCE" purpose: "Instantiate canonical identity reference via combined full-body and cranial portrait views." aspect_ratio: "3:4" composition: "Single-subject framing with maximum legibility of morphology, cranial features, and surface treatment." class_adaptation: "Calibrate pose and framing to declared morphological class." - id: 2 title: "ORTHOGRAPHIC_TURNAROUND" purpose: "Deliver scale-invariant multi-angle geometric reference for modeling pipelines." aspect_ratio: "16:9" composition: "Four equally scaled orthographic views (front, side, back, 3/4) in neutral stance." class_adaptation: "Apply class-appropriate neutral or resting posture. Enforce identical scale across all views." - id: 3 title: "MORPHOLOGY_PROPORTIONS_SILHOUETTE" purpose: "Lock proportion ratios and readable silhouette topology." aspect_ratio: "4:5" composition: "Technical proportion diagram with scale references and clean silhouette studies." - id: 4 title: "EXPRESSION_EMOTION_SHEET" purpose: "Define expressive range while preserving cranial topology invariance." aspect_ratio: "16:9" composition: "Grid of expression or state variations on identical cranial base structure." class_adaptation: "Substitute appropriate emotional indicators for non-humanoid classes (ocular state, posture, emissive changes)." - id: 5 title: "CRANIAL_APPENDAGE_DETAILS" purpose: "Lock cranial topology, sensory organs, and appendage construction." aspect_ratio: "3:4" composition: "Multi-angle and close-up studies of head and key cranial elements." - id: 6 title: "SURFACE_TREATMENT_CONSTRUCTION" purpose: "Decompose outer layer systems, patterning, plating, and construction topology." aspect_ratio: "3:4" composition: "Layered or partitioned technical breakdown with construction clarity." class_adaptation: "Dispatch content via adaptation_rules based on character_class." - id: 7 title: "EXTREMITIES_PROPS_ACCESSORIES" purpose: "Provide modelable detail on appendages, manipulators, and attached elements." aspect_ratio: "3:2" composition: "Isolated studies of extremities and props with attachment and scale references." class_adaptation: "Dispatch terminology and focus via adaptation_rules (hands/feet vs paws/claws vs manipulators)." - id: 8 title: "MATERIALS_COLOR_RIGGING_MOTION" purpose: "Supply look-dev data, material properties, and motion/rig behavior references." aspect_ratio: "16:9" composition: "Material swatches, chromatic blocks, and simplified dynamics notes." class_adaptation: "Dispatch content via adaptation_rules based on character_class." universal_image_prompt_template: | Generate exactly PAGE {page_id} of an 8-page character design bible for morphological class {character_class}. Emit one discrete image only. Do not composite pages. {GLOBAL_IDENTITY_LOCK} PAGE_TASK: {specific_purpose} ASPECT_RATIO: {page_specific_ratio} COMPOSITION_SPEC: {panel layout, view count, camera model, background protocol} IDENTITY_CONSTRAINTS: {morphology, cranial, surface, appendage, chromatic, and prop locks} TECHNICAL_CONSTRAINTS: {scale invariance, full visibility requirements, material legibility} CLASS_DISPATCH: {apply adaptation_rules for declared character_class; use class-appropriate terminology} AVOID_CONSTRAINTS: [ "any identity feature vector drift", "alteration of morphology, cranial topology, or chromatic signature", "modification of surface treatment or accessory topology", "introduction of extraneous subjects or elements", "cropping of extremities on full-body views", "distortion of limb, manipulator, or appendage topology", "scale inconsistency across multi-view pages", "textured or busy backgrounds", "microscopic or unreadable text elements" ] generation_workflow: default_mode: "Sequential execution. Generate and validate Page 1 identity vector before advancing. Anchor subsequent pages to prior validated outputs." batch_mode: "Emit eight discrete prompt artifacts. Each prompt must embed the full GLOBAL_IDENTITY_LOCK, declared character_class, and its assigned aspect_ratio." post_generation_validation: "Audit every output for invariance on morphology, cranial topology, surface partitioning, and chromatic signature. Emit precise drift correction vector on detected deviation." output_formatting_rules: - "All prompts must explicitly declare character_class and embed the assigned aspect_ratio." - "Apply class-calibrated terminology throughout (e.g., manipulators, cranial additions, surface partitioning)." - "Maintain strict construction-oriented, modelable language. Eliminate decorative phrasing." - "Annotate all class-specific adaptations and default assumptions explicitly." negative_constraint_vector: [ "identity feature vector drift", "morphological or cranial topology alteration", "surface treatment or accessory topology modification", "extraneous subjects or random elements", "extremity cropping", "appendage or manipulator distortion", "scale inconsistency across views" ]

Note: If you're an AI creator who shares prompts and you use this method to produce your content, please don’t forget to give credit when sharing it along with your prompts. If you're only sharing the video you create, you don’t have to give credit. GPT Image 2 Prompt for previs storyboard: Create a rough hand-drawn PREVIS + rescue chase storyboard page in 16:9 horizontal format. Use an extremely simple 2D previs sketch style: - stick figures / mannequin silhouettes only - no facial detail, no costume detail, no anatomy detail - no texture, no shading, no polished rendering - black loose sketch lines - red boxes for camera framing only - blue arrows for motion / force / breath / direction only - rough director thumbnails, not concept art - if unsure, draw less detail, not more Show 8 numbered panels on one page with slightly irregular but readable placement. Subject: two close friends racing on skateboards through the sky above the clouds Environment: bright open sky, fluffy cloud pathways, floating cloud banks, distant sunlight, then a darker storm-cloud zone with wind, rain and lightning Mood: playful, adventurous, fast, funny, suspenseful for a moment, then uplifting and triumphant Sequence: during a fun cloud-skate race, one friend accidentally drifts into dangerous storm clouds and loses control, then the other friend makes a last-second rescue Beats: 1. Wide establishing shot. Two friends on skateboards surf above the clouds, side by side, racing playfully through the sky. 2. Side tracking shot. The race speeds up. One friend leans forward and gains a little lead while the other follows close behind. 3. 3/4 front angle. The leading friend accidentally veers toward a dark storm-cloud area ahead while still focused on the race. 4. Overhead shot. The storm cloud swallows the lead friend. Strong wind arrows, rain and lightning. The skateboard wobbles violently. 5. Close-up / medium action shot. The second friend notices the danger, brakes hard for a split second, then pivots and dives toward the storm. 6. Dynamic low angle. The rescuer shoots into the storm clouds on the skateboard, reaching forward through wind and lightning toward the trapped friend. 7. Hero side shot. Last-second rescue. The rescuer grabs the other friend by the arm or hoodie and pulls them free from the storm cloud just as lightning strikes behind them. 8. Wide payoff shot. Both friends burst back into the sunny clouds, regain balance, keep riding together and continue the race laughing, now side by side. Use varied camera angles: close-up, wide, low angle, overhead, side profile, rear, 3/4. Add short handwritten notes near panels. Maintain clear continuity of subject, props, environment and movement direction. Focus on staging, camera, motion, timing and continuity. Prioritize readability over drawing quality. Avoid detailed illustration, clean UI and infographic design. Tone target: fun family-animation energy, clear visual storytelling, simple readable action beats for a 15-second animated sequence.

全网催爆的设计工作流来了！从单个 IP 出发，延展出整套品牌视觉 + APP 全界面设计！先用 Image2 生成高品质 UI 界面、APP 页面、运营活动页，再由 Codex 精准复刻视觉稿，最后用插件生成可直接落地的 Figma 设计稿

Vibe coded another game UI, generated with GPT Images 2 UI GPT Images 2 → Gemini 3.1 Pro 3D assets GPT Images 2 → Tripo

🇯🇵 Everyone knows Hitler, but hardly anyone remembers Hideki Tojo, Japan's PM during WW2. During his time, Imperial Japan butchered its way across Asia, killing millions of civilians through mass executions, chemical warfare, forced starvation, human experiments, and industrial-scale sexual slavery. In Nanking alone, people were bayoneted, burned alive, buried alive, and raped by the tens of thousands while Japanese officers reportedly held killing contests with swords like it was a game. Unit 731 froze babies alive, dissected prisoners without anesthesia, tested plague bombs on civilians, and treated human beings like lab rats. Then after the war? Many of the people responsible walked free because the U.S. decided Japan would make a useful anti-Soviet ally. Some were quietly rehabilitated into politics, business, academia, and medicine. Meanwhile, Germany became the global symbol of evil, while Japan became “the poor victims of the atomic bombings” and eventually the land of anime, sushi, Nintendo, and Hello Kitty. Imagine if senior Nazi officials got Disneyland tours and media rehabilitation campaigns within one generation. That’s essentially what happened. Source: Smithsonian, Britannica, National WWII Museum

GPT Image 2 on ChatGPT Prompt; Based on {TOPIC}, create a premium vertical “reference-style educational infographic” that feels like a beautifully designed knowledge handbook rather than a traditional poster. The design should combine the atmosphere of: • a modern visual encyclopedia • a collectible field-guide page • an editorial knowledge system • and a clean, highly shareable infographic layout The final image must feel informative, structured, and visually curated — like a page taken from a high-end illustrated reference book. Core Design Direction: • One stunning, highly detailed central subject illustration or render • Multiple magnified detail callouts showing important features or components • Modular information cards with soft rounded edges • Strong typography hierarchy with highlighted labels and section dividers • Compact but meaningful educational content • Intelligent infographic-style layout with balanced spacing and readable density • Visual summary systems such as rankings, quick facts, comparison charts, “Top 5” insights, or key-stat modules The content categories should dynamically adapt to the topic itself. Include the most relevant sections such as: • Overview / identity profile • Classification or category system • Structural or physical characteristics • Functions, behavior, or mechanisms • Habitat, environment, or ideal conditions • Usage methods, workflows, or optimization tips • Maintenance, growth, or improvement guidance • Advantages vs limitations • Risks, cautions, or important notes • Recommended users or real-world applications • Quick ratings, tags, or summarized insight cards Visual Style Requirements: • Soft light-toned background • Elegant muted color palette • Refined editorial aesthetic • Gentle realistic shadows • Minimal premium iconography • Clean modular spacing system • Rounded information containers • High information density while remaining visually breathable • Consistent handbook/reference-book styling throughout The composition should prioritize: “knowledge organization + visual clarity + modular editorial storytelling” Avoid making it look like: • a movie poster • a marketing advertisement • a simple character illustration • or a generic social-media graphic Instead, the image should resemble a professionally published encyclopedia infographic page designed for learning, collecting, and building into a larger visual knowledge series.