Lore

Nano Banana Pro

Nano Banana Pro for Pros

The host argues that Nano Banana Pro, run inside Photoshop through the Luminel plug-in, gives photoreal 4K output with multi-reference control. Combined with ordinary Photoshop corrections (blur, curves, masks), it supports full production workflows for portraits, fashion and cinematic lighting.

Kai Weissenfeld · 2025-12-14 · English

Key ideas

  1. The video opens by claiming that Nano Banana Pro's realism and true 4K output mean no online image can be trusted. The host says the model's full potential only opens up through the Luminel plug-in in Photoshop. According to the host, native integration is limited to 1K and cannot take extra images, while the plug-in adds 4K, advanced reference images, a wide range of models and fewer restrictions.

  2. The host claims the model understands prompts well enough that compact or non-verbal inputs can replace long descriptions: GPS coordinates, a photo of a garment, a fashion sketch, a 2D lighting diagram. He shows this across three scenarios: a studio portrait moved into new environments, a fashion sketch turned into a photo, and a lighting sketch turned into a cinematic scene.

  3. The workflow rests on chaining prompts against one image. The subject's face and pose stay consistent while the environment, framing, outfit, era and lighting change. The host presents this as a way to build a set of related frames, for example for a vertical video.

  4. The host treats model output as a strong starting point, not a finished result. Its remaining flaws (missing background blur, a changed outfit, lost facial detail, color drift, grading he doesn't like) are fixed either by re-prompting with the right references or with conventional Photoshop tools. His best results come from alternating generation with manual correction.

  5. In the closing, the host says he is testing Seedream 4.5 inside the same plug-in. He says it might even dethrone Nano Banana Pro, and that a video on it is coming.

  6. Text-only generation (no image input) — Generating an image purely from a prompt by selecting the 'no image input' option instead of supplying a source photo. Apply: Click 'no image input', type a descriptive prompt (for example 'minimal soft focal seascape, hazy gray blue sky and muted blue gray water'), choose Nano Banana Pro and set it to full 4K.

  7. Generating a custom background to use as a reference image — Creating the exact environment you want from scratch at 4K so it can be fed back to the model as a second input. Apply: Generate the seascape, add it as a reference image, turn off that layer's visibility and switch to the flattened document option so the model gets both the subject photo and the background.

  8. Pose-preserving scene change — A prompt that places an existing subject into a new environment while explicitly keeping the pose, with the model adapting the lighting and keeping the face consistent. Apply: With the subject and background references loaded, prompt 'make the woman stand inside the water up to her belly. Make her hair and skin wet. The woman remains in her exact pose.'

  9. Prompt-based lens zoom — Asking the model to 'zoom into' a subject so it re-renders the frame like a longer focal length, with changed depth of field and proportions, instead of digitally cropping. Apply: Type 'zoom into her face' for a portrait, then 'zoom into her left eye, macro close-up' and 'zoom closer in the eye, macro close-up' to go progressively closer.

  10. Google Maps coordinates as location prompt — Pasting GPS coordinates from Google Maps in place of a written location description so the model recognises and renders that exact place. Apply: Copy the coordinates from Google Maps, add a reference photo of the garment and a period such as 'the 1970s', and prompt 'Make her wear the coat and stand straight in this location in the 1970s' followed by the coordinates.

  11. Garment reference image — Adding a photo of a specific coat to the reference images so the model dresses the subject in it while keeping her character consistent. Apply: Add the coat photo to the reference images alongside the subject, then reference 'the coat' in the prompt.

  12. Prompt-based upscaling — Using Nano Banana Pro as an upscaler by prompting it to make a low-resolution picture high resolution. Apply: Feed in the poor-resolution vintage photo, prompt it to 'make the picture high resolution', then zoom in to check the details.

  13. Subject masking and background replacement — Selecting the generated subject with Photoshop's quick selection tool and placing that mask over a different, upscaled background. Apply: Select the entire woman, apply the selection as a mask on the upscaled background, and choose a background whose lighting direction and color mood already match so the bridge position and light line up.

  14. Field Blur with pins for depth of field — Rebuilding realistic depth of field in Photoshop's Blur Gallery by placing several blur pins with different intensities. Apply: Go to Filter > Blur Gallery > Field Blur, place pins across the image, give far areas stronger blur (about 6 px) and near areas subtle blur (about 1 px), and tune until the transition is smooth.

  15. Vertical 9:16 crop — Cropping the finished frame to vertical 9 by 16 so the frames can be used in a vertical video. Apply: Crop the composite to 9:16 before the next restyling step.

  16. Cinematic blue hour night restyle — A prompt that converts a daytime image into a cinematic night blue hour shot while keeping the elements unchanged. Apply: Prompt 'Turn the image into a cinematic night blue hour shot. The elements in the image remain exactly the same. Add some flashlight to the woman.'

  17. Local Gaussian blur patch fix — Correcting an area where the model failed to blur far-away background by blurring only that region in Photoshop. Apply: Select the rectangular unblurred area, copy it to a new layer, apply Filter > Blur > Gaussian Blur and adjust the amount until the patch blends into the background.

  18. Masked curves adjustment for background depth — A curves adjustment layer with a mask that lowers exposure only on the background to add depth and richer blues. Apply: Add a curves layer, mask it so it affects only the background, and lower the exposure for a deeper, moodier, more cinematic look.

  19. Sketch-to-photo conversion — Turning a fashion sketch into a realistic photo that keeps the composition, outfit shape and pose. Apply: Supply the sketch and prompt it to turn the sketch into a real fashion photo taken in Paris with no people in the background.

  20. Background photo as additional reference for campaign control — Adding a specific background photo (a Paris street) via the layer option to control the setting of the generated fashion photo. Apply: Add the street photo as an extra image, switch its layer off again so the sketch is still included, and prompt 'turn the sketch into a realistic fashion photo using the street photo as a background'.

  21. Re-feeding the sketch to correct outfit drift — Supplying the original sketch again as a reference when the generated clothing deviates from the design, while keeping the pose from the photo. Apply: Prompt the model to change the outfit according to the sketch but maintain the pose of the real photo and keep the look realistic.

  22. Mask restoration of lost face detail — Using a Photoshop mask to bring back facial detail that was lost in a later generation. Apply: Mask the earlier, more detailed face back into the corrected image.

  23. Curves color correction for reference drift — Fixing colors that drifted from the original background reference with a curves adjustment. Apply: Add a bit of blue in the highlights and lower the red tone slightly.

  24. 2D lighting-setup sketch to image — Using a 2D top-down lighting diagram, plus a reference image of the actress and a mood/location prompt, to generate a scene matching the lighting layout. Apply: Add the actress as a reference via the layer option and prompt 'Create an image based on the 2D live setup sketch. Use the image of the woman as a reference for the actress. Location, a dark, smokey warehouse. Strong white backlight from a high rectangular window behind her. A flaming barrel on the right.'

  25. Curves, saturation and hue color grade — A post-generation grade that softens contrast and shifts the color mood of the warehouse image. Apply: Add a curves adjustment pulling down highlights and midtones and lifting the deepest blacks, add blue to the shadows, lower saturation slightly and shift hue a bit toward green.

  26. Follow-up zoom after grading — Requesting a closer perspective of the already-graded image, with the model preserving the applied color corrections. Apply: Prompt 'Zoom closer to her face' to get a detailed portrait (skin pores visible) that keeps the grade.

Insights

The video states that a prompt-based zoom differs from a digital crop. Like a longer lens, it changes depth of field and facial proportions: the background and ears go softer and focus lands on the eyes. For that reason the host recommends it for zoom-in effects in video.

Zoom prompts can be chained: face, then left-eye macro, then closer eye macro. The host reports that the last step shows every tiny vein in the white of the eye and a small water droplet on the eyelashes.

The era named in the prompt also changes incidental details. With 'the 1970s' in the prompt, Nano Banana Pro added period cars around the location it recognised from the coordinates, not just a period look on the subject.

The video says a replacement background is easy to composite when its lighting direction and color mood already match the generated scene. Here the bridge's position in the background could be matched as well.

A sharp, upscaled background pasted behind the subject looks unrealistic until depth of field is rebuilt. For the Field Blur pins the host suggests roughly 6 px for far areas and about 1 px for near areas.

The host reports a specific model weakness: after the night-time restyle, a rectangular patch of far background had no blur. He fixed it by copying the patch to a layer and applying a Gaussian blur until it blended in.

To restyle a flattened composite, the host turns off the visibility of the reference layer and uses the flattened-document option. The model then receives two inputs: the studio shot and the separate reference image.

The first sketch-to-photo result with a street background changed the dress too much. The host re-fed the sketch as a reference and told the model to change the outfit per the sketch while keeping the pose of the real photo. He openly treats this correction as part of the workflow.

Color drifted away from the background reference, and the face lost detail. The host restored the face with a mask and corrected the color with a curves adjustment (a bit of blue in the highlights, slightly less red).

The host's grading recipe for the warehouse image: pull down highlights and midtones slightly, lift the deepest blacks to soften contrast, add blue to the shadows, lower saturation slightly and shift hue a bit toward green. A later 'zoom closer to her face' prompt kept this grade and showed skin pores.

The host describes the original studio light setup. It was a white background, two whiteboards on either side of the model, two flash heads aimed into the whiteboards from behind, and a big white reflector bouncing light onto her face. The model later shifted the lighting to match the new water background while keeping the face consistent.

«Don't trust any image you see online anymore.»

— 00:00

«It understands your prompts so well that you can just drop in Google Maps coordinates and it will visualize that exact location just from the numbers.»

— 00:04

«I use Nano Banana Pro directly inside Photoshop with the Luminel plug-in to unlock true 4K resolution and advanced reference image support, something that native integration can't do since it's limited to 1K and doesn't let you add extra images.»

— 00:35

«make the woman stand inside the water up to her belly. Make her hair and skin wet. The woman remains in her exact pose.»

— 03:09

«You don't just want to digitally zoom into the same frame because in real life when you zoom in with a lens, the depth of field and the facial proportions change as well.»

— 03:45

«Make her wear the coat and stand straight in this location in the 1970s»

— 05:42

«Nano Banana Pro is also an amazing upscaler.»

— 06:30

«Objects far away from the camera get the stronger blur, for example, around 6 pixel, while areas closer to the camera only get a subtype blur around 1 pixel.»

— 07:45

«Nano Banana slightly struggles with the depths of field.»

— 08:53

«Nano Banana Pro is already insane, but I'm currently testing Seedream4.5 inside the plug-in, and honestly, it might even dethrone Nano Banana Pro.»

— 13:52

Reception

Mostly positive and impressed by the demo ("stunning", "great demonstration"), but many comments ask about per-image pricing, API keys and installation, and some doubt whether the narrator is real or the demo is staged.

A practical, prompt-by-prompt workflow demonstration that pairs Nano Banana Pro generation with classic Photoshop corrections, and that also promotes the Luminel plug-in. Its value is the order of steps and the specific fixes for model weaknesses (depth of field, color drift, lost face detail).

14:25

↳ Kai Weissenfeld · YouTube

Watch original