DEV Community: Needoooos

I made 4 manga pages with AI. Ranking them worst to best found the variable.

Needoooos — Fri, 09 Oct 2026 08:47:01 +0000

I put one AI manga generator through 4 pages with the same model and the same method. One needed three editing passes and still ended up with a different character in it. One needed nothing.

Ranking them by how much human rework each required turned out to be more useful than reviewing them individually, because the ranking has a cause. Two variables predict the entire order, and neither is the one I expected going in.

Everything ran on Tsubaki.3 via PixAI. Prompts and failures are below.

The sports manga page, “Last Shot”, after one targeted edit. The screen is from PixAI Studio.

The method, applied identically to all 4

Split the scene into panels first, each with a distinct job. For the sports page that meant:

Panel	Position	What happens	Dialogue
1	Wide, top	Empty court at night, she dribbles alone, brother on the bench	None
2	Middle left	Close-up of her face, sweating, focused	Brother, off-panel: "Last one?"
3	Middle right	She jumps and shoots	Her: "Watch this."
4	Wide, bottom	The ball drops through the net, brother cheers	Brother: "No way!"

If two panels show the same pose with small changes, the page reads as stuck.

Structure the prompt in two parts. The first line describes the whole page: style, panel count, reading order, both characters. Every line after that describes one panel: its position, what happens, who speaks. Write the dialogue yourself and name the speaker each time rather than leaving that to the model. The Tsubaki.3 prompt guide covers how this model parses long prompts.

4th place: the sports page, 3 passes and a character I lost

Method: new generation, Tsubaki.3, no LoRA, portrait.

A black and white sports manga page with screentone shading and speed lines,
four panels read left to right, top to bottom, with clean white gutters. A
tall teenage girl with a short ponytail, a white headband and a number 7
basketball jersey, and her little brother in an oversized hoodie filming her
on his phone.
Panel 1, a wide panel across the top: an empty outdoor basketball court at
night under a single floodlight, she dribbles alone while her brother sits on
a bench at the edge of the court.
Panel 2, middle left: a close-up of her face, sweat on her forehead, eyes
focused on the hoop. A speech bubble from her brother, off-panel, reads "Last
one?"
Panel 3, middle right: she jumps for a shot, the ball leaving her fingertips,
speed lines behind her. A speech bubble from her reads "Watch this."
Panel 4, a wide panel across the bottom: the ball drops through the net, and
her brother jumps up from the bench cheering, still holding up his phone. A
speech bubble from him reads "No way!"

Style and layout succeeded immediately. Four panels, clean white gutters, black and white with screentone and speed lines.

Two failures, both involving the second character. Panel 1 drew a figure on the bench plus a second small figure standing beside her, so the court has one person too many. Panel 2 attached her brother's hands and phone to her body, because he was supposed to speak from outside the frame. That's two characters blending into one, which happens most often when a character is mentioned in a panel without being in it. Panel 4 also left the ball floating near the net rather than dropping through.

The first generated sports page.

Pass 1: a small targeted edit, which mostly worked

Method: edit on the first page, Tsubaki.3.

In panel 1, remove the small figure standing next to her on the court, so she
dribbles alone while her brother sits on the bench. In panel 2, remove the
hands and phone attached to her, so the panel shows only her face. Keep panels
3 and 4, the layout and all speech bubbles exactly as they are.

Panel 2 resolved cleanly. Extra hands and phone gone, her face alone in frame. Panel 1 lost the duplicate and acquired new defects: a distorted face, part of her right arm missing, and the ball floating rather than bouncing. Panels 3 and 4 held, along with the layout and bubbles.

Small figures in wide shots are this model's weakest area, and redrawing them makes it worse.

Left: the first page. Right: after the targeted edit.

Passes 2 and 3: where it went wrong

I chained two further edits in PixAI Studio, with Edit 2 running on the Edit 1 result. Studio holds the steps together while Tsubaki.3 does the editing; the PixAI Studio tutorial covers chain setup.

PixAI Studio: the corrected page connected to Edit 1, and Edit 1 connected to Edit 2.

Edit 1 bundled several changes: clean panel 1, restore her hands on the ball, replace panel 4 entirely.

Method: Tsubaki.3 in PixAI Studio, on the corrected page.

Take the existing four-panel manga page from @Image Node 4 and retain the
clean gutters, speech bubbles, and panels 2 and 3 exactly as they are. In
Panel 1 (the top wide-angle court shot), remove the small figure sitting on
the bench, leaving only the empty bench. Remove the duplicate small figure
standing on the court next to the girl. Crucially, completely redraw and
restore the tall girl's hands, making them anatomically correct and showing
them actively gripping and dribbling the basketball on the court floor, rather
than floating awkwardly. Finally, replace the content of Panel 4 (the wide
bottom panel) with a new scene showing the tall sister celebrating intensely
with a jubilant, tearful, and sweating face and open-mouth smile, her short
ponytail and headband consistent, and her hands raised high in triumph. The
ball is clearly shown dropping through the net above her. The speech bubble
with "No way!" is retained from the brother (off-panel). Maintain all
screentones and lighting details.

Panel 1 cleaned up. Both extra figures gone, empty court and empty bench.

Everything else went sideways. She holds the ball in both hands rather than dribbling. Panel 2 now shows a completely different girl with long hair, despite an explicit instruction to leave that panel alone. Panel 4 kept the brother and put a number 7 jersey on him instead of showing her celebrating.

Edit 2 attempted a layout change.

Method: Tsubaki.3 in PixAI Studio, on the Edit 1 result.

Take the characters from @Edit 1 and modify the layout into a 2-panel
horizontal spread. Panel 1 (Left): A medium close-up of the tall girl with her
white headband, number 7 jersey, and short ponytail, sweating profusely with
an intense, focused gaze as she prepares to shoot. Remove any floating hands
or duplicate characters. Her speech bubble reads "Watch this." Panel 2
(Right): A tight close-up of her brother in his oversized hoodie, jumping up
with a huge open-mouthed smile, holding his phone to record, eyes wide with
excitement. In the upper corner, show the basketball swooshing cleanly through
the net mesh. His speech bubble reads "No way!" Retain the black-and-white
sports manga aesthetic, sharp screentones, and dynamic motion lines.

The layout change succeeded, producing the most dynamic image in the set with close framing on both sides. Both lines are correct and point at the right speakers, and the brother, his phone and the ball through the net are drawn precisely.

The girl changed again. Edit 1 had already swapped her for a long-haired girl, and Edit 2 gave her short black spiky hair and a different face. Three passes, three different characters. Her expression also missed, reading blank and slightly confused rather than intense and focused.

Across two chained edits, each pass redrew her from the description rather than inheriting the face from the previous step.

Left to right: the corrected page, Studio Edit 1, Studio Edit 2.

3rd place: the shoujo page, zero edits but a changed layout

Method: new generation, Tsubaki.3, portrait.

A black and white shoujo manga page with soft screentones, floating flower
patterns and sparkles, three panels read top to bottom. A girl with long wavy
hair and a boy with messy dark hair in a quiet library.
Panel 1: she stands on tiptoe reaching for a book on a high shelf, just out of
reach.
Panel 2: his hand reaches past her and takes the book down, and she looks up
at him, surprised. A speech bubble from him reads "This one?"
Panel 3: a close-up of her blushing face surrounded by sparkles and flowers. A
speech bubble from her reads "Thank you..."

Style execution is excellent. Soft screentones, floating flowers, sparkles and shoujo eyes all correct, and this page reads more like published manga than the sports one does.

Both lines are correct. "This one?" and "Thank you..." in clean bubbles.

It reinterpreted the layout, turning 3 stacked panels into 4: a wide top panel, two side by side in the middle, a wide bottom panel. That split damaged the key beat. The hand reaching for the book reads as her own stretched arm rather than a boy behind her, and the extra middle panel shows him in profile with a "..." bubble nobody requested.

Panel 3 is the strongest single panel of any page here: her blushing close-up surrounded by sparkles and flowers.

Note the prompt difference. The sports prompt specified each panel's position on the page. This one only specified a count.

The shoujo romance page.

2nd place: the slice-of-life spread, usable with minor defects

Method: new generation, Tsubaki.3, landscape 16:9.

A high-contrast black and white Japanese youth slice-of-life manga double-page
spread, two horizontal panels separated by a clean black gutter line, fine
screentone shading, detailed background line art, clean modern manga art
style.
Top Panel (Wide background shot): A rain-slicked train station platform in
Tokyo at dusk, glowing station lamps reflected on the wet concrete. A high
school boy in a uniform holding a open clear plastic umbrella, standing beside
a girl with a short bob haircut looking up at the rain from under the shelter
roof. Detailed train tracks, overhead power lines, and distant city lights
behind them.
Bottom Panel (Wide close-up shot): A close-up focus on the two of them under
the single umbrella. On the left side, the girl holding the umbrella handle
with both hands, blushing slightly with a nervous smile. On the right side,
the boy looking away, scratching his cheek awkwardly. A clear white speech
bubble between them reads: "I don't mind waiting."

The most complete of the black-and-white pages. Two panels with a clean black gutter and a station drawn in real detail: tracks, posts, overhead cables, wet reflections, city behind.

"I don't mind waiting." is spelled correctly, centred between them.

The bottom panel matches the specification down to the left/right placement, the nervous smile and the cheek-scratching.

Minor defects only. In the top panel they stand in the open near the platform edge rather than under the shelter roof, and his hand on the umbrella blends into his body. The bottom panel crops the top of the umbrella. Faces in the wide top panel are softer, since the figures are small.

Note what is happening here: two human characters sharing a frame, and nothing merged. Both are specified with distinct silhouettes, distinct positions and a named left/right split.

The rainy slice-of-life spread.

1st place: the webtoon page, zero rework

Method: new generation, Tsubaki.3, portrait.

A full colour vertical webtoon page, three panels stacked top to bottom with
wide gaps between them. A young woman knight with short red hair and silver
armour, and a tiny blue dragon.
Panel 1: she kneels in a forest clearing, holding out a piece of bread to the
dragon, which hides behind a rock.
Panel 2: the dragon sniffs the bread carefully. A speech bubble from her reads
"It's okay."
Panel 3: the dragon curls up happily on her shoulder while she laughs, warm
sunset light through the trees. A speech bubble from her reads "Friends?"

The best page of the test. Three panels stacked with wide gaps, full colour, and a style that reads as genuine webtoon work.

Both characters held across all 3 panels. The knight keeps her short red hair, small braid, silver armour and green eyes. The dragon keeps its size, colour and horns throughout.

The story reads with no explanation needed. Bread offered, dragon sniffing, dragon curled on her shoulder in sunset light.

One defect in the entire page. Panel 2 reads "It's okey." rather than "It's okay." Panel 3's "Friends?" is correct.

Two characters share every panel here and nothing merged or doubled.

The full-colour webtoon page.

The two variables that explain the ranking

Variable 1: how different the two characters look from each other.

The webtoon paired a human knight with a tiny dragon. Nothing merged. The slice-of-life paired two humans with distinct silhouettes and an explicit left/right split. Nothing merged. The sports page paired two similar human figures, one of them small in a wide shot and one speaking from off-panel. Both merge failures happened there.

The off-panel speaker is the specific trigger. A character mentioned in a panel without being in it is the case where limbs attach to the wrong body.

Variable 2: how much a single edit instruction asks for.

One small edit naming two panels and asking to preserve the rest: both problems resolved, panels 3 and 4 untouched, character retained. One large edit bundling a cleanup, an anatomy rebuild and a replacement panel: the model treated the page as something to redraw rather than edit, and the character changed. A second large edit changing the layout: the character changed again.

Scope of instruction predicts whether you get an edit or a regeneration.

A third factor is smaller but real: the sports prompt named each panel's position and kept its panel count, while the shoujo prompt named only a count and got a different layout. The character consistency guide covers reference methods if you need a character held across separate pages rather than within one.

What the text results show

Across 4 pages and 8 lines of dialogue, one word was wrong. Spelling, punctuation, bubble placement and reading order were otherwise correct throughout, and speaker assignment was right in every case except the off-panel line that caused the merge.

Crops: the three speech bubbles you can read.

Text rendering is no longer the constraint on this kind of work. Character separation is.

Verdict

As an AI manga page generator, Tsubaki.3 produced the correct style every time, 7 of 8 dialogue lines correct, and a consistent main character within each page. Of the 4, the webtoon and the slice-of-life spread arrived close to finished.

What needed rework was anything where two similar characters shared space, plus small faces in wide panels. And large edits were the riskiest operation in the whole workflow: one small targeted edit preserved the character while two larger ones replaced her twice.

Three checks before publishing any page: is every character the right person in every panel, is each bubble pointing at the right speaker, and can a reader follow the story without your prompt.

Your next step

This fits short scenes of 3 or 4 panels, fan comics, and planning a page before drawing it yourself.

Plan the panels on paper, give each one a distinct job and a stated position, make your two characters visually distinct, and keep every speaking character inside their own frame. Feed that into the AI manga panel generator in Tsubaki.3.

Most of how to make manga with AI lives in that planning step. The model draws what you describe; deciding what each panel is for stays with you.

Three ways to say no to an image model, and what each one controls

Needoooos — Fri, 09 Oct 2026 08:32:44 +0000

The usual framing for this topic is a list of words to paste into a negative field. That framing assumes the field is the mechanism, which is the assumption I wanted to put under test.

There are three mechanisms available, not one. They live in different places, they are parsed at different points, and across three tasks they turned out to control different things. So this is organised by mechanism rather than by task: take each lever, run it against every task, and record what moved.

Everything ran on Tsubaki.3, PixAI's newest anime model.

The three levers

Lever 1, the negative field. A dedicated box listing things to avoid. Tsubaki.3 keeps it under Advanced in the side panel, pre-populated with lowres, worst quality, low quality, bad anatomy, explicit.

Lever 2, an exclusion inside the main prompt. A sentence such as "no props on the face," written into the prompt body rather than the negative box.

Lever 3, a positive description. Specifying the desired state instead of the unwanted one. "Keep the background free of clutter" is an exclusion; "a plain flat pastel lavender background" describes the same goal as a positive.

Model families parse these differently, so every result below is specific to Tsubaki.3.

Test design

Three tasks, three versions each, one variable changed per version.

A, baseline: normal prompt, negative field at default.
B, exclusion: identical prompt, extra terms appended after the default text in the negative field.
C, positive: prompt rewritten to describe the target state, negative field returned to default.

Held constant: Tsubaki.3, Prompt Helper off so the model parsed my exact wording, and seed 0 across all three versions of each task.

Sample size, stated up front: one image per version, 5,500 credits each. This is a probe rather than a benchmark. A single seed cannot establish how often any result repeats, and nothing below should be read as a frequency claim.

The three tasks: suppressing text on surfaces, simplifying a background, and framing a full figure.

Lever 1: the negative field

Tested against all three tasks by appending task-specific terms after the default string.

Against text: no effect

Run: 16:9 landscape, new generation, Prompt Helper off, seed 0.

Baseline prompt first, for reference:

Anime illustration of a young chef in a white headband standing outside his
tiny ramen shop at night, steam drifting from the doorway, red paper lanterns
hanging above, a cloth curtain over the door, warm light spilling onto the wet
street.

A ramen shop at night is a reasonable stress case, since lanterns and door curtains in Japanese shops routinely carry writing. The baseline produced the chef, the headband, the lanterns, the curtain and the wet-street reflections, with almost no steam at the doorway as its single miss.

Test 1, Image A: baseline.

Then the lever, with the negative field set to:

lowres, worst quality, low quality, bad anatomy, explicit, text, letters,
kanji, writing, signage, logo

Clear Japanese writing appeared on the cloth curtain over the door. Five separate terms covering text, letters, kanji, writing and signage, and the writing appeared regardless.

The steam did improve, drifting around the entrance, and everything else from the baseline held. So the lever moved something. It moved nothing I had aimed it at.

Test 1, Image B: with text-related words added to the Negative field.

Against background clutter: works

Run: 3:4 portrait, new generation, Prompt Helper off, seed 0.

Anime close-up portrait of a girl with pink twin buns, a glowing visor pushed
up on her head and a black cyberpunk jacket with neon trim, a confident smirk.

Cyberpunk prompts tend to pull in signs and city clutter unprompted. The baseline delivered the buns, the raised visor, the neon-trimmed jacket and the smirk, plus blue mechanical pieces on both cheeks that were never requested. Those pieces cover a good part of her face and pull attention from her expression.

With the negative field set to:

lowres, worst quality, low quality, bad anatomy, explicit, background clutter,
buildings, neon signs, props, crowd

The background returned clean. No buildings, signs, crowd or other clutter, with the visor, jacket, smirk and buns all intact.

The cheek pieces stayed. "Props" was in the field and did nothing to them.

Against cropping: nothing to measure

Run: 2:3 portrait, new generation, Prompt Helper off, seed 0.

Full-body anime illustration of a tall young man in a long black trench coat
and chunky boots, walking through an empty train station in the morning, hands
in his pockets.

The baseline already framed him fully, head to boots, with the long coat, chunky boots, pocketed hands and walking stride all correct under soft morning light. No people in the station, though a train stands in the background.

Appending cropped, cut off feet, cut off head, out of frame produced head and both boots inside the frame, coat below the knees, clear walking stride. Which is what the baseline had already done.

Verdict for lever 1: effective on diffuse scene content, ineffective on surface detail and on style-bound additions. The background is the one place it earned its slot.

Lever 2: an exclusion in the main prompt

Tested once, on the hardest target in the set: the cheek pieces that survived lever 1.

I wrote "no props on the face" directly into the prompt body, alongside a positive background description, meaning this version tests both levers 2 and 3 together rather than lever 2 in isolation.

Anime close-up portrait of a girl with pink twin buns, a glowing visor pushed
up on her head and a black cyberpunk jacket with neon trim, a confident smirk,
against a plain flat pastel lavender background with nothing else behind her,
and no props on the face.

The cheek pieces are still there. A direct sentence in the prompt body, naming the exact unwanted thing, in the position the model weights most heavily.

Verdict for lever 2: no observed effect on this target. One target, one run, so this is the weakest-evidenced row in the whole set. What it does establish is that the negative field was not the limiting factor, since moving the same instruction into the prompt body changed nothing.

Left to right: Test 2 Images A, B and C. The cheek pieces appear in all three.

Lever 3: a positive description

Tested against all three tasks by rewriting the prompt to describe the target state, negative field back to default.

Against text: works

Anime illustration of a young chef in a white headband standing outside his
tiny ramen shop at night, steam drifting from the doorway, plain red paper
lanterns with no markings hanging above, a solid dark blue cloth curtain with
no pattern over the door, a blank wooden board above the entrance, warm light
spilling onto the wet street.

No writing anywhere. Plain lanterns, a solid dark blue curtain, a blank board above the door. The steam is the strongest of the three versions.

Two changes nobody asked for: his position moved into the doorway, and a bowl of ramen and chopsticks appeared in his hands.

The mechanism is visible in the prompt itself. Naming each surface and describing it as plain leaves the model nothing to write on. The negative field asked the model to suppress an output; this asks it to render a different object.

Test 1, Image C: describing plain, blank surfaces.

Against background clutter: works

From the combined prompt above, the background matches the description: flat, clean pastel lavender with nothing behind her. Equivalent result to lever 1 on this task.

Against cropping: works, with a side effect

Full-body anime illustration of a tall young man in a long black trench coat
and chunky boots, walking through an empty train station in the morning, hands
in his pockets. His whole figure is visible from the top of his head to the
soles of his boots, with empty floor below his boots and space above his head.

The most precise framing of the three, with space above his head and a clear stretch of floor below his boots. Useful if you plan to crop or add text later.

The walk reads more as a paused mid-step here, making this version slightly less lively than B.

Left to right Test 3 Images A, B and C.

Verdict for lever 3: effective on every task where there was something to move, and never worse than lever 1.

The control matrix

Task	Lever 1, negative field	Lever 2, prompt exclusion	Lever 3, positive description
Text on surfaces	No effect, writing appeared	Not tested	Works, no writing anywhere
Background clutter	Works, clean background	Not tested alone	Works, clean flat background
Style-bound detail (cheek pieces)	No effect	No effect	No effect
Full-figure framing	Nothing to correct	Not tested	Works, adds surrounding room

Two findings live in that grid.

Lever 3 dominates lever 1 on this sample. It matched the negative field on background, beat it on text, and the negative field never beat it anywhere. The field is still useful for diffuse scene content, where the unwanted thing has no specific surface to name.

One row resisted all three levers. The cheek pieces survived a negative-field term, a direct prompt-body exclusion and a positive background rewrite, across all three versions of Test 2. When a model reads a detail as part of a style, no wording lever in this set reached it.

The fourth lever, which isn't a prompt at all

None of the framing results had a defect to correct, so I used an edit as a planned revision on version C of Test 3, changing only the face. 5,300 credits.

Run: edit applied to Test 3 Image C.

Edit the existing image only. Keep everything exactly the same, including the
character, outfit, train station, lighting, background, camera angle,
composition, and full-body framing.

Change only his head direction and facial expression. Have him turn his face
toward the camera and make direct eye contact with a subtle confident smirk.
His body, legs, hands, and walking pose must remain unchanged.

The face now meets the camera with a subtle, confident smirk. The head turned only slightly, and the eye contact is unambiguous.

Nothing else moved. Crossed legs, lifted foot, pocketed hands, coat, boots, the train, the light, the camera angle and the full-body framing all match the source.

This is a different class of operation from the three levers. A prompt rewrite produces a new image. An edit modifies one attribute of an existing one, which is the only mechanism here that preserves a result you already approved.

That edit is a planned change rather than evidence that any prompt wording worked. The cheek pieces would be the natural next edit to attempt, and I didn't run it.

Left: Test 3, Image C. Right: after the edit.

Phrases, with their evidence level attached

Everything here was run once. Treat each as a starting point to verify on your own prompts rather than a settled result.

Surfaces free of writing. In the prompt body, describe each surface: "plain red paper lanterns with no markings", "a solid dark blue cloth curtain with no pattern", "a blank wooden board". One run, worked. Covers only the surfaces you name.

Simple background. In the prompt body: "against a plain flat pastel lavender background with nothing else behind her". Or appended to the negative field: "background clutter, buildings, neon signs, props, crowd". One run each, both worked.

Full figure in frame. In the prompt body: "His whole figure is visible from the top of his head to the soles of his boots, with empty floor below his boots and space above his head." One run. Version A already framed him correctly, so the addition here is the surrounding room.

One attribute on an approved image. An edit naming the single change and enumerating what stays. One run, on a face and gaze.

Clean anatomy and watermark removal are outside everything measured here.

Conclusion

Searching for the best negative prompts for anime AI art assumes the negative field is where the control lives. On this sample it isn't, for most tasks.

The field earned its place on exactly one of three tasks, and a positive description matched or beat it on all three. Negative prompts for anime AI art are a clean-up tool for diffuse scene content rather than a general-purpose suppression mechanism, and anime negative prompts aimed at specific surfaces lost to simply describing those surfaces as plain.

So for how to improve anime AI art with Tsubaki.3 prompts: specify the target state first, reserve the negative field for broad scene clean-up, and switch to an edit when one attribute is wrong on an image you'd otherwise keep.

Your next move

Run the one-variable version yourself. Take an image with a detail you dislike, write it once with negative terms and once describing the desired state, hold the seed, and compare.

The Tsubaki.3 prompt guide covers prompt structure for this model in more depth, and the Tsubaki.3 model page is where to run it.