OD Art: Reliable Working Guidance

Known State documentation rechecked October 30, 2026. Use the actual controls available in your account. These are documented working methods, not a guarantee of identical images or exact preservation. Signed-in availability and live generation performance were not tested in this review.

Jump to the confirmed tag catalog and reported style combinations, added November 2, 2026.

Use the separate Prompt Crafter to assemble character, scene and negative drafts, browse the supplied token library, and keep local presets. Its creator recipes are separate from the confirmed-tag catalog below.

Record the route before changing prompts

Record the selected character, image model, workflow, active appearance fields and presets, scene prompt, and intended change. Distinguish character creation, selected-character Generate, standalone or group generation, in-chat images, and image editing. Do not assume they receive identical inputs.

Check access to the intended control as well as the displayed cost. The public Generate form, retrieved November 1, 2026, labels Custom prompt as Premium and displays an upgrade gate. If that control is locked in your account, stop the custom-text exercise or use an available route; copying prompt text and having enough coins do not unlock it. This public-form observation does not establish every signed-in plan's controls.

Anime character creation changed its default to Fluidity on October 17, 2026. That announcement does not establish a corresponding in-chat model change; check the selected model. Product update

Keep permanent appearance focused

Start with a simple base and the closest useful presets. Body controls interact, so reconcile overlapping presets, Custom Physical Details, and available Face Details. Keep stable face, physique, hair, and markings there; move temporary clothing, pose, expression, and location to the scene. Correct conflicting definitions rather than universally blanking fields or appending another corrective block. Build the appearance

Match wording to the model

Dreamy and Fluidity usually favor short visual phrases; Vivid and editing models can use natural descriptions. Use underscores for exact known Fluidity tags and ordinary spaces for other phrases. Describe the visible result, group related traits, and remove redundant or contradictory cues. Begin unweighted; adjust emphasis cautiously after resolving wording and field conflicts. Quality/scoring slogans and exact body measurements are unreliable controls. A word-count suggestion is a starting point, not a universal cap. If the current route exposes a dedicated Negative Prompt field, use it for unwanted visual elements instead of adding negative wording to the positive prompt. Image prompting reference

Confirmed tags and reported style tests

Added November 2, 2026. The following 37 tags were supplied as user-confirmed, alongside observations from an artist reference. The original test model, route, settings and test date were not included. These observations describe the reported results; they have not been independently retested for this update.

Confirmed tag catalog

Choose the treatment you need from the catalog. Keep the character and scene fixed while comparing styles.

The descriptions “ink wash,” “woodblock print” and “comic dots” explain the preceding tags. In collage_(medium), the parentheses are part of the supplied tag.

Reported combinations and interactions

The expressions below preserve the supplied notes. “Plus” and “with” identify cues used together; they do not establish a special prompt operator. Keep reported weights within their recipes rather than treating them as universal defaults.

Watercolor and soft anime

In the reported comparison, watercolor was weak on its own and barely registered over a strong artist influence. The source recommends sumi-e for ink wash; ink wash and watercolor remain different treatments. It also records this combination for a soft anime look:

(((Watercolor_style))) with sketched_lineart:1.2

Flat 2D anime

Reported as the strongest flat 2D treatment in the supplied comparison:

anime_style:1.3 plus flat_color:1.3 plus cel_shading

The catalog contains cel_shaded; this recipe uses cel_shading. Preserve both spellings rather than assuming they are aliases.

Black-and-white manga

Reported to produce black-and-white manga linework:

junji_ito plus monochrome, screentone

Sepia interaction — unsuccessful

sepia plus monochrome

The supplied test reports an incorrect-looking result, with monochrome not fully overridden. No successful sepia override was supplied.

Ligne-claire

Reported to produce a clean European comic treatment:

ligne-claire, clean-linework, muted-color-palette

Additional supplied ink-and-line recipe

This separate example was supplied without an accompanying result or selected model. Its complete wording is retained for comparison:

score_9, score_8_up, score_7_up, (kentaromiura:1.2), (monochrome:1.1), (greyscale:1.0), (lineart:1.2), [(sketch:1.1), (hatching(texture):1.2), (crosshatching:1.2), (ink_drawing:1.1), (heavy-inking:1.1)], high_contrast, masterpiece, best_quality, highly detailed:1.1, [subject]

Replace [subject] with the visible subject and scene you want. The score tags and quality phrases are retained as part of this example, with no confirmed result attached to them here. They are not added to the confirmed catalog or made general requirements. Preserve the grouping when reproducing the supplied example, and check the selected route before relying on its bracket or weighting syntax. No universal numerical meaning for the parentheses or square brackets is established by this recipe.

Reuse without losing the comparison

Start a fresh comparison with a simple, unweighted prompt. When reproducing a reported weighted recipe, retain its exact terms and record the model, route, appearance inputs and resulting image. Keep anime distinct from anime_style:1.3, and watercolor distinct from (((Watercolor_style))). Inspect identity and scene preservation separately from whether the intended rendering appeared.

Supply the information the route needs

Selected-character Generate already receives permanent appearance; supply scene changes such as outfit, pose, setting, lighting, and temporary expression. Keep these changes out of permanent appearance fields. Create the gallery

For a standalone or group prompt, include the necessary visible identity for each subject. Keep each person's traits, clothing, pose, and position attached to that person. Names alone are insufficient when the destination does not supply their definitions.

Translate story language into visible language

The chat LLM and an image workflow have different immediate tasks. A sentence such as Mira is worried that the parcel will arrive late leaves the picture unspecified. Choose a visible moment without claiming that its appearance proves an internal feeling. This October 31 authoring exercise does not establish image-model performance.

For a selected-character Generate scene, after confirming that the route supplies Mira's permanent appearance, an original scene prompt could be:

Mira wearing a navy jacket, standing beside an archive counter, looking at a
sealed parcel on the counter, three-quarter view, warm indoor light.

The character's careful vocabulary belongs in APD; the jacket, pose, objects, framing, and light describe this picture. For a standalone route, add the necessary visible identity rather than expecting the name Mira to supply it. For a group, attach each outfit and position to its subject instead of writing one loose shared list.

Practice one revision: replace a tense, meaningful delivery scene with a specific arrangement of subjects, objects, and visible actions. Then inspect whether that arrangement appears. A mood word may complement those facts, but cannot substitute for them. Keep a requested style change separate from a change to the character's face or body.

Edit from a clean in-app image

Choose an existing generated image as the editing source. State the requested change and which identity, proportions, pose, crop, clothing, background, and lighting should remain. Work on one major category at a time. Inspect the result against the accepted source; if identity drifts, return to a good base and simplify. Reassess long edit chains. Save successful images with their prompts, using Remix or More → Details when available. Preservation is an acceptance check, not a promise. Image consistency and editing

Do not assume external reference-photo upload is available for character/image generation. Upload policy

Where offered, brush selection can target an edit area; inspect the entire result for unintended changes. The September 5, 2026 update added this control to regular image editing and removed the video seed control. That removal does not establish still-image seed availability. Editing update

Run one small comparison

This is an original practice procedure, not a measured result. Use an original adult character and check the displayed cost before every generation.

1. Save a simple accepted image and its appearance fields, selected model, route, scene prompt, and exposed settings.
2. Write one target, such as change the navy jacket to dark green. List what should stay recognizable: face, hair, build, pose, and framing.
3. Change only that scene instruction, or use the accepted image as an Edit source. Do not mix those routes in one comparison.
4. Score the jacket change separately from unwanted identity or scene changes. Record an obscured feature as not assessable, not a pass.
5. Keep failures and repeat within your chosen budget before adopting the wording. Return to the accepted source if an edit drifts.

Expected result: a saved input/output pair and a specific judgment about the intended change and preservation. A successful example is not evidence that the next output will match. Use the test log to retain the observations.

Keep experiments separate

A withdrawn field-placement workaround

Supplied community material reviewed October 31, 2026 explicitly retracts the claim that persistent details placed in a skin-tone field automatically appear or disappear correctly under clothing. The author reports that other factors affected the initial examples and clean-character tests contradicted the explanation. The retraction itself has no supplied date. Keep it as a correction, not as an active skin-tone recipe or proof that every incidental result must fail.

Use the documented appearance fields for stable identity and inspect clothing coverage and visible marks in the actual image. A hidden mark is not a failure merely because clothing covers it. A claim about one in-chat route does not establish the same behavior on Generate or Edit. The current-form guide resolves changing appearances without assuming a hidden field override.

Interpret creator edit reports narrowly

The supplied discussion reports multi-pass scene edits, imperfect pose and door preservation, and later external splicing and audio work. No complete controlled input/output series was supplied. Keep each accepted source, requested change, failed attempt, crop, charge, and returned result. A successful character likeness does not establish correct room geometry or a stationary camera.

The abbreviation PD is not expanded in those notes; do not silently map it to Public Description or CPD. Artist/style names can also appear as unwanted text in a reported result. Use descriptive rendering language and inspect the actual route rather than inferring a universal placement order. The suggestion to substitute vapors for smoke or mist has no observed result in the supplied text; it remains a lexical experiment.

Historical artist/style experiments remain source material, not active defaults. No fixed weight budget, guaranteed BREAK separation, exact nested-weight arithmetic, or automatic artist recognition has been established. Adopt a cue only after repeatable improvement on the same recorded route without unwanted identity or scene changes. See the source review.