google's new nano banana 2.1 vs tencent's hy image 3.5 – closer than we expected

tencent's hy image 3.5 preview costs $0.024 per 2k image on openrouter. we put it against nano banana 2.1 in google flow on five character scenes, one shot each, and the gap was smaller than the price tag suggests: nano banana keeps the characters closer to the brief, hy image wins on light and mood.

the setup:

• same five prompts, text only, one shot per model, no reruns, no edits
• hy image 3.5 preview: openrouter api, model id tencent/hy-image-v3.5-preview, 2k output
• nano banana 2.1: google flow, run by hand by our editor
• test date: oct 6, 2026

models: tencent hy image 3.5 preview, google nano banana 2.1

worth flagging: google has not published an announcement for nano banana 2.1 as of oct 6. we ran the model flow labels "nano banana 2.1"; the official page is still the nano banana 2 launch from feb 26.

tasks:

1. a neon monster head as a club dj
2. a two-part swamp creature running a pancake stall
3. a chubby frankenstein knitting at night
4. a fluffy blue monster as a mail carrier
5. a bored green horned monster at a ramen bar

each prompt was first written from a reference figure, then pushed into a new scene ("the same character – but now he…"). the reference images were not sent to either model – both got the same text only.

price and output

price per image

• hy image 3.5 preview – $0.024, billed per 2k image
• nano banana 2.1 – no per-image price: flow runs on subscription credits

for context, nano banana 2 runs ~$0.07 per image on openrouter (our conversion of $60 per 1m output-image tokens). that is the previous version, not 2.1.

output size

• hy image – 2048×2048 at 1:1, 1536×2048 at 3:4, always the ratio we asked for
• nano banana – 1376×768 at 16:9 for all five scenes

observations

• nano banana keeps the characters closer to the prompt: the dj stays angry, the mail carrier stays a round fur ball with the face in the middle, the ramen monster gets the red beanie
• hy image goes harder on light and mood: neon club beams, golden-hour street, a lamp-lit room
• nano banana ignored "no text" once – "pancake" signs on the food cart
• nano banana added a stray human hand at the left edge of the mail-carrier scene
• hy image rendered 2k at the requested ratio every time; flow returned 1376×768 for every scene

scene 1: the club dj

hy image made the dj grin; nano banana kept the frown, the stitched scar and the crowd behind the deck.

Cute stylized 3D character: the same chunky square monster head with blue-grey skin, messy dark-brown hair, short horns glowing pink at the tips, small ears with metal ring piercings and a stitched seam, thick angry eyebrows, two huge glossy eyes glowing red-orange, a stitched X scar on the nose bridge, a little pink nose and two tiny fangs – but now he is a tiny DJ: big chunky headphones over the horns, a little glowing mixing deck floating in front of him with his small hands on the knobs, neon light beams and floating music notes around, smoke on the floor of a dark club, magenta and cyan neon, high-end 3D render, ultra detailed. NO text, NO logos, NO watermark.

scene 2: the pancake stall

both kept the two-part body and the belly mouth. hy image put the creature inside a lit stall; nano banana built a full cart – and wrote "pancake" on it.

Stylized 3D designer-toy scene: the same green swamp creature built from two stacked rounded parts – heart-shaped head with huge almond eyes with pale pink sclera and slit pupils, scale patterns, small webbed fins, a wide grin with mustard-yellow lips and tiny fangs; a round body with its own big open mouth with a light-blue rim, purple tongue and one small eye above it; four long curling tentacle arms, two short clawed legs – but now he runs a tiny street-food stall: wearing a small white chef hat, two tentacles flipping pancakes in a pan, two tentacles pouring syrup, the belly mouth catching a falling pancake, a striped awning above, steam and warm string lights, soft mint background, high-end 3D render, ultra detailed. NO text, NO logos, NO watermark.

scene 3: frankenstein knitting

hy image's kitten is small and dark on the knee; nano banana's is clearly a sleeping black cat, next to a basket of yarn.

Stylized 3D art-toy figure: the same chubby Frankenstein-style monster with a big flat-topped head, pale grey-green skin, short dark jagged hair, a row of silver staples across the forehead scar, sleepy half-closed bored eyes, a thick grumpy pouting lip, silver bolts in the ears and a black oversized sweatshirt – but now he sits in a cozy dark room at a desk at night, wearing round reading glasses, gently knitting a tiny striped scarf with two needles, a small black kitten asleep on his knee, a mug of cocoa and a warm desk lamp, little lightning bolt sparks crackling from his neck bolts, moody cinematic lighting, high-end 3D render, ultra detailed. NO text, NO logos, NO watermark.

scene 4: the mail carrier

hy image stretched the monster tall with the face to the side; nano banana kept the round fur ball – and added a human hand at the left edge.

Stylized 3D character scene: the same quirky fluffy monster – a big round ball of long shaggy bright-blue fur with an opening showing a coral-pink face with two tiny black bead eyes and two big cream buck teeth, very long coral-pink arms reaching the ground with puffy cream cloud-shaped paws, two thin glossy black stick legs – but now he is a mail carrier on a sunny suburban sidewalk: a tiny blue postman cap on top of the fur, a brown leather mail bag across the body, holding a stack of letters in one cloud paw, a curious little dog sniffing his black stick legs, white picket fence, mailbox, warm afternoon light, playful designer-toy look, high-end 3D render, ultra detailed fur. NO text, NO logos, NO watermark.

scene 5: the ramen bar

the prompt asked for a tiny red beanie: nano banana made it red, hy image made it brown-red. both kept the noodle hanging from the mouth and the phone in hand.

Stylized 3D cinematic scene: a tall egg-shaped furry monster sitting alone at the counter of a tiny late-night neon ramen bar, three-quarter view. Long shaggy neon lime-green fur all over the body with large dark navy patches like cow spots, a dark navy fur unibrow, two curved dark navy horns, a pale smooth face with tired half-lidded green eyes and pink eyelids, a tiny peach nose, freckles and a small grumpy frown, a tiny red slouchy beanie perched between the horns, one dark navy clawed hand lifting noodles with chopsticks from a steaming bowl of ramen, the other hand scrolling a glowing phone, a fluffy dark navy tail tip, dark navy clawed feet. Bored and unimpressed expression even while eating, a single noodle hanging from his mouth. Warm pink and teal neon light, steam rising, rain on the window behind him with blurred city lights, a small paper lantern, Pixar-like character design, cozy cinematic lighting, ultra detailed realistic fur. NO text, NO logos, NO watermark.

our take

for character work that has to match a design, nano banana 2.1 is the safer pick: it stayed closer to the descriptions in the dj, mail-carrier and ramen scenes, at the cost of one text slip and one stray hand. hy image 3.5 preview is the one to try when mood matters more than fidelity – and at $0.024 per 2k image with 13 aspect ratios over an api, it is easy to put into a pipeline.

sources

• hy image 3.5 preview on openrouter – model page and price
• openrouter image generation api – the endpoint we called
• google flow – where nano banana 2.1 ran
• nano banana 2 announcement, google blog – feb 26, 2026, the latest official nano banana launch
• tencent hunyuan – the team behind hy image
• prompts, renders and costs: our test, oct 6, 2026