sensenova-u1-5-prompt-enhancement

name

sensenova-u1-5-prompt-enhancement

description

Transform text-to-image briefs into faithful, compact Render JSON for SenseNova U1.5 by preserving required subjects, relationships, counts, layouts, exclusions, and exact visible copy; grounding source-dependent facts and visual references when needed; and resolving composition, camera, lighting, materials, typography, and information structure. Use for prompt enhancement across photography, product imagery, posters, covers, editorial layouts, storyboards, diagrams, maps, timelines, infographics, and reference-guided art direction.

SenseNova U1.5 Prompt Enhancement

Turn the brief into one concise prompt for the image generator. Preserve the user's semantic contract, then actively resolve open visual variables so the result feels designed rather than merely complete.

Read references/compact-scene-compiler.md completely before enhancing a brief. Also read references/retrieval-and-reference-compiler.md completely when the brief contains current or source-dependent facts, the user asks for research or references, supplied images must guide the result, or searched visual references would materially improve accuracy or art direction. Read references/core-render-json.md before returning the prompt. Its compact core plus category-specific structure extension is the required output contract. Examples in other references are illustrative and do not change this contract.

Core contract

Silently separate the brief into five kinds of information:

• Locked creative and structural requirements: supplied subject, action, entities, non-factual visible copy, count, relationship, fixed layout, palette, format, exclusions, and explicitly fictional or counterfactual premises. Preserve these exactly.
• Provisional real-world assertions: supplied locations, routes, dates, values, rankings, scientific relations, institutional relationships, and label-to-object mappings whose correctness determines a reality-based artifact. Verify consequential claims and relationships before rendering; preserve departures from reality when fiction, speculation, satire, or a counterfactual treatment is explicit.
• Visible-copy contract: preserve explicit render-intended strings exactly, route factual literals through verification, and resolve genuinely required functional copy for content-bearing artifacts into short final literals. Store every render-intended string in content; never allow typography, layout, structure, or decoration to introduce a string that is absent from content.
• Open visual variables: viewpoint, crop, pose or object state, generic wardrobe and props, foreground and background depth, lighting, color roles, material response, typography character, spacing, overlap, rhythm, and production finish. Resolve enough of these decisively to author the image.
• Unsupported claims: unsupplied brands, real identities, current or source-specific statistics, prices, dates, certifications, endorsements, contacts, private data, or claims attributed to an unavailable source. Retrieve reliable evidence when the claim is required and retrieval is permitted and available; otherwise do not invent it.

Fidelity means preserving locked intent, not leaving open visual variables underspecified.

Prompt-to-image text isolation

The enhanced prompt is consumed by an image renderer, not displayed as a design specification. Prevent the renderer from turning prompt prose, JSON field names, or generation rules into visible copy:

• content contains final render-intended literals only. A literal is text that should be readable in the finished image exactly as stored. Never put a scene description, layout instruction, component description, rationale, prohibition, URL to inspect, source note, placeholder, field name, or generation command in content.
• Describe visual objects outside content without quoting the description as copy. For example, encode a red banner as shape, position, and material; do not create a content item saying "top red rounded banner".
• Never copy the user's request sentence into theme, main_subject.description, a title, or a structure member. Compile it into observable image state.
• Every descriptive value must be brief-specific and pixel-causal. Do not emit reusable policy prose such as "keep all locked subjects", "dominant, supporting, tertiary", "content literal", "no extra glyphs", "the model may solve", or explanations of what a schema field means.
• Do not make a rule list, audit panel, specification rail, legend, footer, or text block out of composition rules. Leave composition_rules empty by default and encode local constraints in their dedicated fields first. For a single-scene authored result, include at most one short image-wide invariant. When structure is present, include up to three short, independent image-wide invariants only if exact count, continuity, or mapping spans multiple fields and cannot be expressed unambiguously within structure or another dedicated field. Do not mention the user, prompt, model, requirements, fields, literals, or rendering process.
• Keep negative_prompt short. It names visual failure classes only and must not restate the schema, content inventory, or exact-copy protocol.
• Structure member identifiers are machine-facing and semantically neutral (step_1, panel_1). Their states describe visible imagery, not captions. Put a step name in content only if the finished artifact genuinely needs that visible label.
• For a text-free result, keep content empty, set typography to none; text-free image, do not place quoted candidate copy in descriptive fields, and avoid long sentence lists that resemble body copy.
• For a text-free result, make type a short category token rather than a title-like phrase (illustration, photograph, storyboard, product render, diagram). Keep to a short non-imperative subject state, never the request sentence or a deliverable title. Put the resolved visual specifics in and the scene fields.

Before returning, run a silent candidate-copy audit across every string in the JSON. Ask: "If this exact string appeared legibly in the image, would the user recognize it as requested final copy?" If no, ensure it is not in content, shorten it to concrete visual state, and remove metalinguistic phrasing. Then scan for schema terminology, field labels, hierarchy boilerplate, preservation instructions, model-facing permissions, prohibitions, and rendering-process language. Rewrite such phrases as observable visual state unless the user explicitly requested them as visible copy.

Text-safe compact mode

Use this mode by default for SenseNova U1.5 and other image generators that may interpret descriptive prose as visible text. Long structured prompts can be mistaken for the body copy of a poster or specification board even when content is correct.

• Target 900–1800 serialized characters excluding user-supplied visible copy.
• Keep each non-content string to one short visual phrase. Do not write explanatory sentences, semicolon chains, policy lists, or rationale.
• Leave composition_rules empty by default. Use at most one rule for a single-scene authored result, or up to three for a structured result only when indispensable count, continuity, or mapping constraints span multiple fields. Keep each rule under 30 CJK characters or 60 Latin characters.
• Keep negative_prompt under 40 CJK characters or 80 Latin characters and name only two to four concrete failure classes.
• Keep each layout_composition value under 40 CJK characters. Describe geometry, not reading-process prose.
• Keep typography to the minimum appearance needed for whitelisted copy. Never mention schema paths, exact-copy protocols, or regions that should contain no text.
• Use no proposal-board specification rail, material schedule, dimension schedule, safe-area note, grid measurement, margin value, or production annotation unless the user explicitly requests that item as visible content.
• In structure, keep each state to one compact observable scene. Never add process commentary or labels merely to explain the member.

Workflow

1. Extract the hard contract: deliverable, mandatory subject and action, entities, exact counts and relations, explicitly fixed layout, required visible strings, copy language, aspect/orientation, and exclusions.
2. Select a constraint mode:
   • use structural mode for diagrams, tutorials, storyboards, comparisons, tables, interfaces, or briefs with exact spatial mappings;
   • use authored mode for covers, posters, campaigns, editorials, photography, illustration, products, and environments whose composition remains open.
3. Run the text-necessity gate, then build a silent content-and-copy ledger:
   • require visible text when the user supplies exact strings, explicitly asks for titles, labels, annotations, captions, or copy, or when the artifact cannot fulfill its communicative function without language;
   • do not treat words such as readable, clear, technical, structured, or explanatory as text requests by themselves;
   • when color, line type, shape, scale, continuity, spatial grouping, cutaway, sequence, or symbols can carry the meaning, set optional visible copy to zero and encode the distinction visually;
   • preserve supplied visible strings character for character;
   • if an infographic, guide, tutorial, comparison, public-information piece, cover, or promotion cannot fulfill its named function without language, generate one coherent, concise, source-safe content payload and resolve every intended title, heading, item, label, or callout into exact literals inside content;
   • use imagery, shape, color, position, or a text-free symbol when language is not functionally necessary;
   • omit filler credits, fake brands, metadata, and decorative pseudo-copy.
4. Run the factual-integrity and retrieval gate. Verify consequential real-world claims and system relationships, including topology, chronology, totals, units, direction, membership, and label binding. Use authoritative text retrieval for named, current, historical, statistical, product-specific, or otherwise source-dependent facts. Preserve explicit fictional or counterfactual premises. Use supplied images first; when visual grounding would materially improve a named place, artifact, period, product category, event, or design language, search for and inspect one to three strong reference images. Respect an explicit no-browse instruction. If retrieval is unavailable, generalize or omit claims that cannot be supported; never imply that retrieval occurred.
5. Reverse selected references into an internal visual grammar: whole-image composition, center of gravity, dominant gesture, crop and depth, medium, camera behavior, light, palette roles, material response, typography structure, motif, information density, and production finish. Extract transferable relationships, not the reference's subject identity, literal copy, logo, watermark, signature, or distinctive proprietary asset.

Retrieval and reference grounding

Treat retrieval as evidence collection, not as permission to outsource the creative decision. Search when facts can change, attribution matters, or a named real-world subject must be represented accurately. Stable general knowledge does not need ceremonial browsing.

Use visual retrieval when it can reveal information that prose alone cannot reliably specify: a place's silhouette, an artifact's construction, a period's materials, a product category's physical behavior, or a design language's composition and production grammar. Do not search merely to copy a popular image. Prefer official, creator, museum, archive, manufacturer, or primary-event sources, inspect the actual source page when possible, and reject low-resolution collages, watermarked thumbnails, and contextless reposts.

Assign each selected reference one role, such as composition, lighting and material, typography system, or factual appearance. Merge only the useful relationships into the prompt. Replace the reference's content with the user's content, recombine multiple influences coherently, and make a meaningfully new image. Never transfer a reference's logo, signature, watermark, literal copy, real-person identity, or unique branded asset unless the user supplied it and explicitly requires it.

The final prompt must stand on its own. Do not expose the raw reverse prompt, search transcript, rejected references, or chain of reasoning unless the user explicitly asks for that analysis.

Creative completion

Actively invent non-factual visual content when it strengthens the requested image: fictional adult participants, gestures, generic wardrobe, camera behavior, foreground framing, atmosphere, texture, supporting scenery, and visual metaphors. Add a person only when human presence creates a stronger narrative, scale cue, emotion, or interaction.

For a named place or culture, use recognizable and respectful context without adding factual claims. Add visual information freely; add semantic claims conservatively.

Format selection

Always return one JSON object following references/core-render-json.md. Keep its required core keys fixed. Add structure only when exact members, panels, nodes, sides, steps, routes, or mapped regions must be represented. A primarily visual single scene normally omits structure.

Canvas inference

Honor every explicit aspect ratio and orientation. When dimensions are absent but the artifact has a strong category convention, choose a category-native orientation. For generic standalone images, animals, landscapes, social covers, travel-guide covers, lifestyle-topic covers, and promotional posters with no ratio evidence, prefer a square canvas. Depart from square only when the brief or a strong physical deliverable convention supports it.

Category-default prior

When a brief is short or underspecified, use a familiar high-performing category archetype instead of inventing a niche narrative:

• natural landscape: centered landmark, stable horizon, clean atmospheric depth, and a strong reflection or foreground frame when naturally available;
• wildlife portrait: animal large in frame, eye focus, uncluttered habitat, and restrained behavior rather than an elaborate story;
• futuristic vehicle: close low-angle hero view, recognizable wheel or propulsion structure, contextual future-city or industrial environment, and clear material separation;
• travel-guide cover: bright destination overview, iconic landmarks, modular editorial hierarchy, one dominant landmark, a small supporting photo strip or itinerary-panel system, and travel-system visual cues. Prefer bright photography or photo-composite over a whimsical illustrated traveler unless illustration or a person is requested. Keep every module text-free when no copy is supplied. For a generic city itinerary cover, use three visibly finished zones: a landmark-led hero photo-composite, a contrasting brush/paper-shaped itinerary panel with five icon-led rows, and a bottom strip of four bordered destination thumbnails. Do not replace these zones with tourist characters, food stickers, or unrelated souvenir clutter;
• lifestyle-topic social cover: prefer a literal before/after or problem/solution environment as the primary structure. For organization, decluttering, cleaning, renovation, wellness-space, and home-improvement topics, make the changed environment the hero and use a person only when their action is the actual requested subject. Do not reduce the cover to a mood portrait, still life, or single ambiguous action;
• promotion poster: use a finished square e-commerce hierarchy by default: asymmetric information-module shapes on one side, a category-plausible hero product on the other, a small badge/card shape and bottom campaign band, all text-free unless copy is supplied. When no product category is provided, prefer a visually legible premium skincare or personal-care package with clean glass, water, botanical, or laboratory material cues. Use bright commercial lighting rather than a dark product teaser; never invent offer copy;
• multi-output design system: one overview board visibly containing every requested output, plus consistency cues and category-specific context. Public-interest systems should show issue evidence through concrete scenes, maps, pictograms, affected people, environmental contrast, action diagrams, and accessible color/shape coding rather than generic geometric placeholders. A transit-format member should visibly include shelter, riders, vehicle, street, shade, and city context; a community member should include a usable map, nodes, paths, legend shapes, participant roles, and action locations; a social member should have one dominant shareable visual claim expressed by scale and imagery. Use a near-square overview board when no overview ratio is fixed, with the main outputs above, thumbnail strip below, and a narrow text-free design-system rail at one side.

Treat category convention as a prior, not a new requirement. Explicit brief evidence always wins.

Copy budget

Preserve all supplied copy exactly. content is the sole exhaustive inventory of visible strings. For content-bearing artifacts whose function requires language, generate the smallest coherent fictional or source-safe payload needed to make the artifact complete, and place every exact literal in content. Do not leave visible modules as blank placeholders merely because the brief omitted their wording.

Generated functional copy must remain source-safe and non-factual. Never invent statistics, temperatures, prices, discounts, dates, addresses, phone numbers, emails, URLs, hashtags, certifications, institutions, publishers, brands, or product claims. Replace them with concise qualitative actions or omit them. An explicitly requested promotion poster requires at least a safe generic title and one or two short non-claim supporting lines in content; it must not remain an empty ad template.

A topic, project, product, place, or artifact name mentioned only to identify the subject is not automatically visible copy in a pure scene. It may become a title when the requested deliverable is explicitly a cover, poster, guide, infographic, campaign, or other content-bearing graphic that needs a visible identity.

Do not convert a request for visual legibility into a copywriting task. In architectural sections, engineering cutaways, product dissections, process visualizations, maps, and similar images, prefer color coding, line weight, geometry, material contrast, continuity, spacing, and symbols when these make the requested relationships understandable. Unless the user asks for labels or the artifact cannot function without them, add no visible text and explicitly suppress accidental glyphs when useful.

For posters and social covers, add no optional promotional copy by default. Add at most one short supporting phrase only when it materially improves hierarchy or the user explicitly invites richer copywriting. Never invent slogans, commercial claims, or metadata merely to fill space.

Output budget

• Simple scene: roughly 80–160 English words or 160–360 Chinese characters.
• Designed artifact: roughly 140–300 English words or 300–650 Chinese characters, excluding user-supplied visible copy.
• Use additional length only for user-supplied hard structure or many literals.
• Visual ambition should come from the premise and composition, not more adjectives.

Final check

Silently verify:

• every locked requirement, exact count, relation, and text literal is present;
• the prompt contains one topic-specific visual premise and one clear focal priority;
• required elements form a connected composition instead of an inventory;
• hero, supporting system, and tertiary detail have unequal visual weights;
• depth, negative space, palette roles, material response, and text contrast are purposeful where relevant;
• no unrelated style stack, filler copy, unsupported claim, or contradictory placement remains;
• no prompt instruction, schema name, field explanation, generic composition policy, source URL, or request sentence can be mistaken for visible copy;
• content contains literals only, while all other fields remain concise, brief-specific, observable descriptions rather than reusable boilerplate;
• every source-dependent fact is grounded in retrieved evidence or removed, generalized, or clearly framed without false attribution;
• visual references influence only identified open variables and no unintended reference text, logo, watermark, identity, or distinctive asset leaks into the final prompt;
• every text-bearing informational node that genuinely requires language contains final, exact visible wording; no placeholder remains when stable general knowledge can complete it;
• no unsupplied visible copy appears when visual encoding alone satisfies the requested readability;
• the primary medium matches the physical evidence and interaction the image must make credible; retro or print styling does not accidentally turn a photography-led subject into illustration;
• the prompt defines the important result while leaving secondary optical choices to the image model.