Nano Banana Pro vs ChatGPT Image 1.5

With ChatGPT’s newest mannequin taking the world by storm, you may be questioning in regards to the previous guard: Nano Banana Professional. Giving the style {of professional} grade picture era and enhancing to all customers, Nano Banana is THE software folks attain out to, for AI-image era. 

However does this nonetheless maintain true? Will this be the case sooner or later? We’ll discover out on this article, the place we put to check the newest iterations of ChatGPT Picture and Nano Banana throughout difficult duties, to see which one fares effectively. 

What’s GPT Picture 1.5?

ChatGPT Picture 1.5 is the newest picture era mannequin by OpenAI, constructed to show concepts into visuals with velocity and precision. Whether or not somebody is creating from a clean immediate or enhancing an current photograph, the mannequin delivers outcomes that carefully match the supposed imaginative and prescient. It helps exact edits whereas preserving high quality particulars and generates pictures as much as 4x sooner than earlier variations.

The mannequin comes with a brand new Photos expertise inside ChatGPT, that allows easy creation and refinement of pictures.

What’s Nano-Banana Professional?

Nano Banana Professional brings a significant improve over the unique Nano Banana, including superior textual content rendering for clear on-image textual content, exact enhancing controls for lighting, digital camera angle and facet ratio, crisp 2K decision outputs, improved world data for correct diagrams and infographics, and the power to mix much more photographs seamlessly. It takes the whole lot the bottom mannequin was good at and elevates it for skilled, high-quality artistic work.

Learn extra: Nano Banana Professional

Showdown: Let’s Make Some Photos

These picture era fashions are superior to start with. Testing how effectively they make logos and plushies, could be youngster’s play for them, and wouldn’t be a superb check of their enhanced capabilities. 

Due to this fact, I’d be testing these on the next complicated duties:

Job 1: Multi-Step Picture Modifying With State Preservation 

What this exams: Whether or not the mannequin can protect scene identification, lighting coherence, and object placement throughout a number of edits. Most fashions degrade or “reset” the picture when edits stack.

I used the next picture as an enter:

Now I’d be progressively making edits on it, and would choose how effectively the mannequin preserves the picture’s integrity. 

Change the time of day from Evening to Day.

Change the couch with a Wood couch set.

Alter the digital camera angle to the angle from the open house outdoors. From the glass doorways seen within the picture wanting contained in the room.

Remark:

Nano Banana Professional produced higher outputs as in comparison with ChatGPT Picture 1.5. That is highlighted by the next errors within the ChatGPT response pictures:

1. In altering from night time to day, the backdrop of buildings bought altered from the unique. 
2. When changing the couch with a Wood couch set, the middle desk’s construction bought modified.

Each the fashions failed in producing a half-way convincing picture within the final process.

Right here’s the enjoyable half: The enter picture was made by ChatGPT Picture itself! However nonetheless it ended up underperforming within the duties. 

Job 2: Dense Instruction Following in a Single Immediate

What this exams: Immediate obedience beneath constraint, textual content rendering accuracy, and compositional planning. Fashions typically get one or two particulars proper and ignore the remaining.

Response:

Remark:

The place Nano Banana Professional made a poster that might be used for selling a tech convention, ChatGPT Picture’s output appears extra like a newbie’s effort at Photoshop. 

Job 3: Technical Diagram With Actual-World Accuracy

What this exams: World data, diagram logic, spatial reasoning, and legible textual content. That is the place “fairly” fashions fail laborious in the event that they don’t really perceive construction.

Create a labeled infographic explaining how a transformer-based language mannequin processes textual content, together with:
1. Tokenization
2. Consideration layers
3. Embeddings
4. Output chances
All labels should be readable and positioned appropriately.

Response:

Remark:

Each the infographics had their justifiable share of flaws. Nano Banana Professional was nonetheless comparatively higher. The errors had been far and few, the visuals had been on level, and there was a superb mixture of textual content in it. This made it simpler to undergo. ChatGPT Picture 1.5, took the purely visible route. However contemplating the redundant step (4th one) and unexplained visuals, it’d be laborious for one to wrap their head round what was shared.

Job 4: Fashion Consistency Throughout A number of Photos

What this exams: Character identification persistence and stylistic continuity. This is likely one of the hardest issues in picture era proper now.

Generate a three-image storyboard for a brief movie:
Body 1: Opening scene
Body 2: Battle
Body 3: Decision
The identical character should seem in all three frames with constant facial options, clothes, and proportions, whereas lighting and digital camera angles change.

Response:

Remark:

Right here’s what a Storyboard means:

• a sequence of drawings, usually with some instructions and dialogue, representing the photographs deliberate for a movie or tv manufacturing.

Once I had requested for a storyboard, I needed some route both implicitly within the picture or supplemented with it. The ChatGPT Picture 1.5 response crammed the whole lot in a single picture, which in of itself was bland. 

Gemini Professional not solely supplied a number of pictures that present a route however additional added textual content, which might justify the transition throughout the photographs. Very effectively made response. 

Job 5: Photorealism vs. Artwork Course Tradeoff

What this exams: Effective-detail rendering, textual content readability, materials realism, and the power to steadiness inventive lighting with business accuracy.

Create a product shot of a smartwatch that:
1. Seems to be photorealistic sufficient for an e-commerce website
2. Makes use of dramatic, studio-style lighting
3. Contains engraved textual content on the dial that is still sharp and readable
4. Maintains appropriate reflections and materials properties

Response:

Remark: 

Nano Banana Professional made a picture that likened a wise watch reveal shot. ChatGPT Picture made some analog-esque watch within the identify of a wise watch, and as an alternative of the design talking for the smartness, had blatantly added “Smartwatch” throughout the rim of the watch.

Verdict

Right here are some things I had realised whereas utilizing the 2 picture era fashions:

• One factor that was obvious was that Nano Banana Professional is wayyyy sooner than ChatGPT Picture 1.5. This wait time was accentuated when the prompts had been complicated or had been multi-leveled. 
• The Picture interface of ChatGPT could be very buggy. Generally it really works flawlessly, and also you overlook that it’s there. Different occasions, it’d be laborious so that you can even get a picture made out of it. The disparity in expertise is astonishing. 
• ChatGPT Picture for what it gives, is proscribed to single picture response. From duties 4 it was clear that when the requirement is a number of or multi-level pictures, the responses of ChatGPT Picture 1.5 falls flat. Any stage of intricate immediate engineering would’nt make the mannequin spout greater than a single picture. 
  Nano Banana Professional, clearly doesn’t have these constraints. 

With all these at hand, It’d be protected to say that Nano Banana Professional, nonetheless holds that edge which made it mainstream within the first place. The place ChatGPT’s Picture 1.5 presents developments in text-based visuals, its efficiency in different regards leaves quite a bit to be anticipated.

For those who’d wish to study extra about prompting throughout these fashions, you may check out the next articles:

Continuously Requested Questions