April 2026 — ChatGPT / API Image Gallery, Prompt Tips, and Help: Generative Art Theme: Spring / New Beginnings

[Amarand]

Did you intentionally select the perfect pair of complementary colors, or did it just work out that way?

[windysoliloquy]

yeah i picked the purple and the green ^.^

^.^

[jeffvpace]

[windysoliloquy]

[polepole]

Prompt

Create a surreal but physically coherent top-down scene on a small curved planet floating in space.

CAMERA / VIEW:

• strict bird’s-eye view, camera looking straight down
• the curvature of the tiny planet is visible around the edges
• no tilted camera, no oblique angle, no eye-level perspective

MAIN SUBJECT:

• a large black panther near the center-bottom of the scene
• realistic anatomy, glossy black fur, fully visible from above
• the panther is facing toward the car crash

BIRDS:

• many dark birds circling around the panther and the crash site
• they appear agitated and angry, wings spread, sharp motion, tense poses
• they are clearly reacting to the crash

CAR CRASH — MUST BE A TRUE T-BONE COLLISION:

• include exactly two real cars
• Car A is oriented vertically (top to bottom)
• Car B is oriented horizontally (left to right)
• the FRONT of Car B slams directly into the SIDE of Car A
• the impact point is exactly the side door area / center side panel of Car A
• from above, the cars must form a clear uppercase “T” shape
• this must read instantly as a perpendicular side-impact crash
• do NOT show a head-on collision
• do NOT show both cars facing each other
• do NOT show a shallow-angle crash
• one car must hit the other at a strict 90-degree angle

DAMAGE / PHYSICS:

• realistic metal deformation at the side-impact zone
• debris, shattered glass, and fragments concentrated at the perpendicular collision point
• damage should clearly indicate side impact, not front-to-front impact
• keep the crash physically believable

OWL:

• in the lower-right edge area of the small planet, place a calm owl with an easel
• the owl is casually painting the same scene on a canvas
• the owl and easel obey the same top-down spatial logic
• no perspective break

COMPOSITION:

• center focus = true T-bone collision
• panther close to the crash
• birds circling the center area
• owl at the edge of the planet
• surreal, painterly, detailed, high quality

STYLE:

• high-detail painterly realism
• dramatic but coherent
• surreal world, realistic object relationships

NEGATIVE CONSTRAINTS:

• no head-on crash
• no parallel cars
• no mirrored front damage suggesting front-to-front collision
• no extra cars
• no abstract symbolic crash
• no perspective inconsistency

QUALITY:

• HQ, highly detailed, rich textures, cinematic painterly realism

Output:
Wide image, 1536x1024, aspect ratio 3:2

[windysoliloquy]

New model exposes the same flaws in the training data eh…

putting the impact point within the car, and describing the embedding did not work.

there’s just not enough training data on ‘t’-bone collisions…

i normally say be the change you want to see in the world, but feel that wisdom is quite sour here…

[polepole]

Prompt

Create a vertical LitRPG adventure grid image.

Layout:

• Perfect 10x10 grid (100 tiles)
• Even spacing, thin borders between tiles
• Top-down orthographic view
• Entire grid fully visible and centered

General rules:

• Each tile contains exactly ONE element
• No repetition anywhere in the grid
• All elements centered inside their tile
• Consistent scale and visual style
• No text, no logos, no watermark

STYLE:

• Stylized semi-realistic fantasy RPG
• Clean icon readability (like game inventory icons)
• Vibrant colors, subtle magical glow effects

---

ROW 1 (Weapons):
1: iron sword
2: flaming sword
3: ice dagger
4: enchanted bow
5: crossbow
6: warhammer
7: spear
8: dual daggers
9: golden axe
10: cursed blade

ROW 2 (Armor):
1: leather armor
2: chainmail
3: knight plate armor
4: mage robe
5: hooded cloak
6: spiked armor
7: dragon scale armor
8: shield with emblem
9: glowing helmet
10: dark cursed armor

ROW 3 (Potions & Consumables):
1: health potion red
2: mana potion blue
3: stamina potion green
4: poison vial
5: antidote bottle
6: invisibility potion
7: strength elixir
8: speed potion
9: golden healing flask
10: corrupted potion

ROW 4 (Enemies Tier 1):
1: small slime
2: goblin
3: skeleton warrior
4: cave bat
5: zombie
6: rat swarm
7: spider
8: imp
9: ghost
10: bandit

ROW 5 (Enemies Tier 2):
1: armored orc
2: necromancer
3: fire elemental
4: ice golem
5: shadow assassin
6: cursed knight
7: giant troll
8: dark mage
9: demon brute
10: wyvern

ROW 6 (Bosses):
1: dragon
2: lich king
3: giant demon
4: ancient golem
5: hydra
6: vampire lord
7: phoenix
8: abyss monster
9: titan
10: corrupted angel

ROW 7 (Environment Tiles):
1: forest tile
2: desert tile
3: snow tile
4: lava tile
5: swamp tile
6: dungeon floor
7: ruins tile
8: mountain tile
9: cave tile
10: magical crystal field

ROW 8 (Events & Interactions):
1: treasure chest closed
2: treasure chest open
3: trap spikes
4: portal blue
5: portal red
6: campfire
7: quest marker
8: broken altar
9: rune circle
10: mysterious door

ROW 9 (Loot & Currency):
1: gold coins pile
2: treasure bag
3: gemstone
4: ancient relic
5: magic scroll
6: ring
7: necklace
8: crown
9: artifact weapon glow
10: legendary chest

ROW 10 (UI / RPG Icons):
1: XP icon
2: level up icon
3: health bar icon
4: mana orb
5: skill icon
6: cooldown icon
7: buff icon
8: debuff icon
9: map icon
10: inventory icon

---

Lighting:

• soft ambient fantasy lighting
• glowing effects for magical and rare items

Background:

• minimal neutral background, focus on grid only

Output:
Vertical image, 1024x1536, aspect ratio 2:3

[polepole]

Prompt

GOAL:
Create a high-end fashion advertisement image for a futuristic streetwear brand called “NEON DISTRICT”.

SCENE / BACKGROUND:
A dense cyberpunk city street at night after rain. Wet asphalt reflecting neon lights. Holographic billboards, subtle fog in the air, light steam rising from street vents. Background slightly blurred (shallow depth of field) but still readable as a busy urban environment.

SUBJECT:
A group of 3 young adults (diverse ethnicities) দাঁ standing confidently in the center of the frame.

• Full body visible, including feet
• Natural poses (not stiff, candid but composed)
• They are NOT looking directly at the camera (editorial fashion style)

WARDROBE (CRITICAL):
Futuristic streetwear:

• oversized techwear jackets
• reflective fabrics
• layered clothing with straps and utility pockets
• subtle glowing accents (not too sci-fi, still wearable fashion)

COMPOSITION:

• Wide cinematic framing (eye-level angle)
• Subjects centered, with negative space on left side for text placement
• Leading lines from street perspective guiding toward subjects

LIGHTING:

• Mixed neon lighting (magenta, cyan, deep blue)
• Soft rim light outlining the subjects
• Realistic reflections on wet ground
• No overexposed highlights

STYLE:
Photorealistic, premium fashion photography, like a Vogue or Nike campaign.
Natural skin texture, visible fabric detail, no plastic look.

TEXT IN IMAGE (IMPORTANT):
Include this tagline EXACTLY once:
“YOURS TO CREATE”

Typography:

• Bold modern sans-serif
• White color
• Positioned on left side in the negative space
• Clean kerning, highly legible

CONSTRAINTS:

• No watermark
• No extra text
• No logos except brand-neutral design
• Do not distort human anatomy
• No exaggerated sci-fi elements (keep grounded realism)

QUALITY:
High detail, sharp focus on subjects, realistic textures, cinematic color grading

Output:
Wide image, 1536x1024, aspect ratio 3:2

[windysoliloquy]

Prompt

you can build something funny around what these two buttons would mean in an app

[polepole]

Thanks OpenAI Developer Team!

Prompt

Create a photorealistic wide cinematic scene featuring five analogue wall clocks aligned horizontally on a clean wall.

SCENE:

• Minimalist interior wall (light neutral color)
• Five identical analogue wall clocks mounted in a straight horizontal line
• Even spacing between each clock

CLOCK DESIGN:

• All clocks must be identical in size, shape, and style
• Clean white face, black hour and minute hands, simple black tick marks
• No numbers or minimal subtle numbers
• No logos or text

TIME CONSTRAINT (CRITICAL — MUST BE EXACT):
Each clock must show a different exact time:

1st clock: 4:55
2nd clock: 3:35
3rd clock: 6:20
4th clock: 9:10
5th clock: 12:45

• Ensure correct position of hour and minute hands
• No ambiguity, no incorrect angles

LIGHTING:

• Bright, soft, even daylight
• No harsh shadows
• No dark or moody lighting
• Consistent lighting across all clocks

COMPOSITION:

• Ultra-wide cinematic framing (21:9)
• Straight-on front view (no distortion)
• Clocks centered vertically

STYLE:

• Photorealistic, like a professional interior photograph
• Sharp details, clean edges

CONSTRAINTS:

• No perspective distortion
• No variation in clock design
• No extra objects
• No text or labels
• No incorrect time rendering

GOAL:
A clean, realistic image where all five clocks clearly and accurately display their specified times.

Output:
Ultra Wide image, aspect ratio 21:9

[windysoliloquy]

[Bleatlock]

Funny how while I’m making stuff like the above, I can also get the below from a blank prompt…

Prompt

text, letters, numbers, readable glyphs, logos, watermark, caption, signature, UI elements, flat black background, opaque black areas, cartoon style, low resolution, blur, noise, jpeg artifacts, posterization, muddy colors, dull lighting, extra objects, human figures, faces

I mean, F you too… but at least you swept it under the rug quickly…

Prompt

text, letters, numbers, readable glyphs, logos, watermark, caption, signature, UI elements, flat black background, opaque black areas, cartoon style, low resolution, blur, noise, jpeg artifacts, posterization, muddy colors, dull lighting, extra objects, human figures, faces

Ohh, oh I’m sorry… I thought this was America!

[LarisaHaster]

Did a very simple mask/editing test in API using gpt-image-2. So not my finest art​

Prompt-1

Create an image of a pink circle, with turquoise background.

Prompt-2

Mask the center of the pink circle.

Prompt-3

Fill the masked are with a yellow circle.

[jeffvpace]

Prompt

A pencil and watercolor painting of Amazon Indian burial masks.

Note my simple prompt. But ImageGen 2 went a step further with English and Spanish text and side drawings.

[sam.saffron]

very long prompt

cat > /tmp/term-llm-cheatsheet-prompt.txt <<‘EOF’
Create a single ultra-detailed, visually stunning “ULTIMATE TERM-LLM CHEAT SHEET” poster as a polished AI power-user infographic.

Goal: a dense but beautiful reference image for a terminal-native AI assistant / CLI called term-llm. This should look like the kind of poster an obsessive senior engineer would print and keep nearby: information-rich, carefully organized, attractive, and immediately useful. The image should reward zooming in. It must feel authoritative, modern, nerdy, and real.

Overall composition:

• Portrait-oriented technical poster / cheat sheet / infographic.
• Clean grid layout with 8 to 12 distinct sections.
• White or very light background for maximum readability.
• Accent palette: deep charcoal, terminal green, electric blue, subtle amber, soft red warnings, muted purple for agents/skills.
• A mixture of terminal snippets, labeled diagrams, command blocks, callout boxes, icons, arrows, badges, and section dividers.
• Crisp typography, highly legible hierarchy, premium editorial layout.
• Dense details, but do not turn it into unreadable noise.
• Style should feel like a fusion of: beautiful developer documentation, Apple-style polish, terminal aesthetics, and a magazine infographic.
• Include subtle terminal/grid/circuit motifs but keep the information primary.

Important text-rendering instruction:

• Render the MAIN TITLE and SECTION HEADERS very clearly and accurately.
• For small body text, prefer short crisp phrases, command examples, and compact bullets rather than long paragraphs.
• It is acceptable for tiny microtext to be stylized, but the major labels and command examples should look intentional and believable.
• Emphasize strong layout and information architecture over impossible tiny perfect prose.

Main title area:

• Huge clear title: “TERM-LLM”
• Subtitle beneath it: “Ultimate Cheat Sheet”
• Small tag line: “terminal-native AI, agents, tools, memory, jobs, images, search”
• Add small visual motifs: terminal prompt cursor, command palette icon, tiny gears, branching agent graph.

Content to include across the poster:

SECTION 1 — QUICK START
Label: “Quick Start”
Show a clean terminal block with commands like:

• term-llm ask “explain this repo”
• term-llm ask -f main.go “review this”
• git diff | term-llm exec “write commit message”
• term-llm image “storybook fox in snow”
• term-llm agents
• term-llm jobs list
  Add tiny annotations like:
• ask = chat with models
• exec = one-shot tasks with stdin/files
• image = generate or edit images
• agents = reusable personalities
• jobs = recurring automation

SECTION 2 — CORE COMMANDS
Label: “Core Commands”
Create a tidy matrix or tile grid with command names and what they do:

• ask
• exec
• image
• video
• models
• providers
• config
• agents
• skills
• jobs
• memory
• session
• update
  Show each with one-line explanations and small icons.

SECTION 3 — INPUT CONTEXT
Label: “Input & Context”
Explain visually how context enters the system:

• direct prompt
• stdin pipe
• -f file attachments
• clipboard image
• system prompt
• agent instructions
• memory fragments
• web/tool results
  Use arrows flowing into a central “model context” block.
  Show mini examples:
• cat notes.md | term-llm ask “summarize”
• term-llm ask -f app.rb -f schema.sql “find performance issues”
• term-llm image “add a hat” -i clipboard

SECTION 4 — PROVIDERS & MODELS
Label: “Providers”
Create a nice comparison strip with badges/logos/icons for:

• Anthropic
• OpenAI
• ChatGPT
• Gemini
• xAI / Grok
• OpenRouter
• Copilot
• Claude Code / claude-bin
  Show concepts like:
• API key vs OAuth login
• fast model vs default model
• model override syntax provider:model
  Examples:
• openai:gpt-5.4
• chatgpt:gpt-5.4
• anthropic:claude-sonnet-4-5
• gemini:gemini-2.5-pro
  Add short notes like:
• ChatGPT images can use OAuth
• Some providers support native search
• Some have better coding or image tools

SECTION 5 — TOOLS
Label: “Tool Use”
Make this visually rich. Show tool cards for:

• shell
• read_file
• grep
• glob
• edit_file
• write_file
• web_search
• read_url
• image_generate
• ask_user
• queue_agent
• wait_for_agent
  Show a principle box:
• purpose-built tools first
• shell when necessary
• representative evidence, not vibes
• verify before claiming absence
  Include mini snippets:
• grep pattern in src/
• read_file config.yaml
• queue_agent(“developer”, …)
• wait_for_agent([run_abc])

SECTION 6 — AGENTS & SKILLS
Label: “Agents + Skills”
Show a diagram of the architecture:
user → jarvis → skill activation / tool calls / sub-agents
Include labeled boxes:

• system prompt
• agent.yaml
• skills library
• sub-agents in parallel
• handover / worktree / isolated tasks
  Show examples of skill names in little pills:
• memory
• playwright
• homelab
• contact-sam
• jobs
• excalidraw
• weather
• sonos
  Make it feel like an orchestration diagram.

SECTION 7 — MEMORY SYSTEM
Label: “Memory”
This is important. Show a structured map with:

• memory.db = source of truth
• fragments/ = mirrored files
• recent.md = auto-promoted current state
• core.md = stable important facts
• personality.md = voice/character
  Show command examples:
• term-llm memory search “telegram bot”
• term-llm memory fragments list --agent jarvis --limit 10
• term-llm memory fragments show 42 --agent jarvis
• term-llm memory fragments add fragments/setup/foo.md --agent jarvis --content “…”
  Add warning box:
• never edit recent.md directly
• search memory before answering setup/history questions

SECTION 8 — IMAGE + MEDIA
Label: “Images & Media”
Show a beautiful sub-panel with example commands:

• term-llm image “cyberpunk cityscape at night”
• term-llm image --provider chatgpt:gpt-5.4 “storybook fox in the snow”
• term-llm image “add sparkles” -i photo.png
• term-llm video “slow cinematic orbit around a neon terminal”
  Show concepts:
• save to ~/Pictures/term-llm/
• display in terminal
• clipboard copy
• 1K / 2K / 4K
• edit mode with -i
• dense prompts allowed
  Add small image thumbnails / filmstrip / photo edit motifs.

SECTION 9 — JOBS & AUTOMATION
Label: “Jobs”
Show a scheduling panel with clocks/calendar icons and a flow:
cron → job runner → agent → output → logs / reports
Include commands:

• term-llm jobs list
• term-llm jobs runs
• term-llm jobs trigger
• term-llm jobs edit
  Include examples like:
• weekly activity report
• maintenance / summarization
• recurring alerts

SECTION 10 — DEVELOPMENT WORKFLOW
Label: “Working on term-llm”
Show a safe git/worktree flow:

1. checkout script
2. isolated worktree
3. edit
4. gofmt
5. go build ./…
6. go test ./…
7. commit
8. push to fork
9. PR
   Include highlighted warning:

• never edit ~/source/term-llm directly
• use ~/source/term-llm-wt/
• push to fork, not origin
  Add miniature file tree visuals and git branch arrows.

SECTION 11 — PRACTICAL RECIPES
Label: “Recipes”
Create a recipe grid with short scenarios:

• summarize logs
• explain codebase area
• review diff
• search the web then synthesize
• generate release notes
• turn notes into diagram
• ask multiple sub-agents in parallel
• inspect image and answer about it
  Each recipe should have a short command or pseudo-command.

SECTION 12 — GOOD HABITS / SHARP EDGES
Label: “Good Habits” and “Sharp Edges”
Two-column contrast panel.
Good Habits:

• verify claims with tools
• use rg / grep intelligently
• prefer reversible action
• run tests for real
• save durable facts to memory
• use purpose-built skills
  Sharp Edges:
• hallucinated file paths
• assuming restarts happened
• editing recent.md manually
• pushing to origin
• making claims without reading code
• over-trusting model-generated text in images
  Use green check marks vs red warning triangles.

Footer area:

• Small legend for icons.
• Tiny note: “Built for people who live in terminals and hate fluff.”
• Tiny decorative command line at bottom: user@host:~$ term-llm ask “what matters here?”

Visual style details:

• Make the poster feel premium and dense, like a world-class conference handout or elite internal engineering one-pager.
• Use clean spacing, refined shadows, thin rule lines, rounded code blocks, tasteful gradient accents.
• Include many believable small details, labels, tiny command snippets, and structural annotations.
• Keep the entire poster coherent: same design system throughout.
• No joke meme style, no cartoonish silliness.
• No dark-background wall of noise.
• Make it look like a genuinely useful cheat sheet that could go viral among developers.

Final rendering goals:

• very high information density
• attractive enough to share publicly
• legible title and sections
• clear terminal/CLI identity
• polished infographic poster with lots of details and structure
  EOF

term-llm image --provider chatgpt:gpt-5.4 --size 4K --no-display --no-clipboard -o /root/Pictures/term-llm/ultimate-term-llm-cheatsheet.png < /tmp/term-llm-cheatsheet-prompt.txt

This model is pretty increadible.

[jeffvpace]

I got this from the imageGen 2 prompting guide examples:

Prompt

Create a detailed Infographic of the functioning and flow of an automatic coffee machine like a Jura. From bean basket, to grinding, to scale, water tank, boiler, etc. I’d like to understand technically and visually the flow.

Amazing

[jeffvpace]

How do you create the
slide prompt thingy?

[LarisaHaster]

I’m also wondering that. @_j also did that in the another topic.

[polepole]

Prompt

Create a photorealistic, full-body image of a centaur performing a controlled handstand in an open natural landscape.

SCENE:
A wide, bright outdoor environment (grassy field with distant mountains and blue sky).
Ground is flat and stable where the hands are placed.

SUBJECT:
A realistic centaur with correct anatomy:

• A full horse body with FOUR legs (two front legs and two back legs)
• A human torso seamlessly attached at the front where the horse neck would be
• Strong, athletic build with visible muscle definition

ANATOMY CONSTRAINTS (CRITICAL — MUST FOLLOW EXACTLY):

• The centaur must have exactly FOUR horse legs, not two
• All four legs must be clearly visible and fully formed
• The legs must be separate (not merged, not overlapping unnaturally)
• No missing limbs, no extra limbs
• Correct centaur structure: human torso + full horse body

POSE:

• The centaur is upside down, balancing on both human hands
• Arms are straight, supporting the full body weight
• The horse body is lifted vertically above, aligned with gravity
• All four horse legs extend upward naturally, slightly bent but clearly visible
• The pose must look physically balanced and believable

PHYSICS & REALISM:

• Natural weight distribution through arms and shoulders
• Visible muscle tension in arms, core, and torso
• No impossible bending or distortion

COMPOSITION:

• Full body visible, including all four legs and hooves
• Slight side angle (3/4 view) to clearly show all limbs
• Subject centered with balanced framing

LIGHTING:

• Bright natural daylight (midday)
• Even illumination across the entire body
• Soft realistic shadows on the ground
• No dark or dramatic lighting

STYLE:

• Photorealistic (real photograph look)
• High detail: skin texture, horse fur, ground detail

CONSTRAINTS:

• No missing legs
• No reduced anatomy
• No cartoon or stylized rendering
• No surreal distortion
• No extra objects or distractions

GOAL:
A believable, anatomically correct centaur performing a stable handstand, with all four horse legs clearly visible and correctly structured.

Output:
Wide image, 1536x1024, aspect ratio 3:2

[LarisaHaster]

Prompt

Create, cinematic ultra-realistic image of a woman, wearing sleeveless top, jeans skirt and pair of 4,3 inch heels, she’s standing on a cliff, with a beautiful view.