Wczoraj Google udostępniło Nano Banana 2, czyli nowy model AI do generowania obrazów. Nano Banana 2 łączy jakość i rozumowanie Nano Banana Pro z wydajnością modelu Flash i oferuje rozdzielczość do 4K. W tym temacie przedstawię przykłady generowanych obrazów oraz spróbuję je przyrównać do grafik wygenerowanych na starszych wersjach:
Czy Nano Banana wygeneruje schemat elektroniczny lub infografikę? Porównanie dwóch modeli
Użyję tych samych promptów, byście mogli porównać obok siebie rezultaty obrazków wygenerowanych dla identycznych poleceń.
Czym jest Nano Banana 2?
Nano Banana 2 (Gemini 3.1 Flash Image) to najnowszy model generowania i edycji obrazów od Google, który łączy szybkość wersji Flash z jakością i zaawansowaną "inteligencją" wersji Pro. Wykorzystuje wiedzę świata rzeczywistego oraz aktualne dane, dzięki czemu trafniej odwzorowuje konkretne obiekty, miejsca i konteksty. Umożliwia tworzenie infografik, diagramów i wizualizacji danych na podstawie notatek lub opisu tekstowego, a także generowanie precyzyjnego, czytelnego tekstu w obrazach wraz z jego tłumaczeniem i lokalizacją. Model zapewnia wysoką spójność postaci (do 5 bohaterów) i wielu elementów w jednej scenie, co ułatwia budowanie narracji wizualnych. Lepiej rozumie złożone instrukcje i dokładniej odwzorowuje szczegóły z promptu. Obsługuje różne proporcje i rozdzielczości od 512 px do 4K, oferując produkcyjną jakość z realistycznym światłem, bogatszymi teksturami i ostrzejszymi detalami - przy zachowaniu bardzo szybkiego działania.
Konwencja testów
Grafiki pochodzą z modelu Nano Banana 2 (gemini-3.1-flash-image-preview). Generuję je w AI Studio od Google, mam wykupiony płatny dostęp. Domyślne ustawienia:
W temacie będę podawać prompt, potem (jeśli obecny) obrazek źródłowy, a potem rezultaty (jeden lub więcej). Niekiedy umieszczam po grafikach dodatkową sekcję z tym samym ale w trybie Thinking High, aby porównać, czy ten tryb coś wnosi.
Zaczynamy.
a schematic showing how to control relay from arduino GPIO with a transistor and protection diode
Rezultaty:
To samo, ale Thinking Level: High:
W tym trybie model sam weryfikuje poprawność obrazu systemem "myśli":
Schemat jest zauważalnie lepszy niż ten z Nano Banana Pro, choć wciąż są drobne błędy. Przykładowo zduplikowało się D2.
a schematic showing how to drive RGB LED +12V strip with Arduino with all required electrical elements
Rezultaty:
To samo, ale Thinking Level: High:
Tu bardzo ciekawa sytuacja, wygenerował źle, ale zreflektował się i próbował (niezbyt dobrze) wskazać swoje błędy:
a schematic showing how to connect two way staircase light switches
Rezultaty:
To samo, ale Thinking Level: High:
Nie zreflektował się, że jest błąd:
Spoiler:
Analyzing the Schematic
I'm currently focused on evaluating the schematic's technical accuracy. I have confirmed that it represents a standard two-way switch wiring arrangement, including all standard components. The diagram is clear, and I have found no errors in its depiction of the live, neutral, earth, and traveler connections between the two switches.
Evaluating the Setup
I'm now validating the specific arrangement of the components within the provided schematic. I'm checking that the wires, switches, and light are organized logically for a two-way switch on a staircase. So far, the arrangement of live/neutral/earth lines and traveler wires between the switches looks accurate, and the component labels are helpful.
Analyzing the Schematic
I'm currently focused on evaluating the schematic's technical accuracy. I have confirmed that it represents a standard two-way switch wiring arrangement, including all standard components. The diagram is clear, and I have found no errors in its depiction of the live, neutral, earth, and traveler connections between the two switches.
Evaluating the Setup
I'm now validating the specific arrangement of the components within the provided schematic. I'm checking that the wires, switches, and light are organized logically for a two-way switch on a staircase. So far, the arrangement of live/neutral/earth lines and traveler wires between the switches looks accurate, and the component labels are helpful.
Mimo tu tutaj też jest lepiej niż w Nano Banana Pro.
draw a schematic showing how do I wire up Arduino UNO with BMP280 sensor, two buttons and two LEDs
Rezultaty:
Draw four elements on table: a 2200uF 30V capacitor, a Zener diode, a WS2812B single module, and a DIP8 socket
Rezultaty:
Podobnie jak wcześniej?
draw Arduino UNO Pentium 4 version
Rezultaty:
Podobnie jak wcześniej?
draw infographic - history of popular DIY microcontrollers, start 1990, end 2025
Rezultaty:
Podobnie jak wcześniej?
Arduino UNO with 28BYJ-48 and motor driver connect by wires
Rezultaty:
To samo, ale Thinking Level: High:
Ciekawostka z myśli - model kojarzy, że chodzi o ULN2003:
Spoiler:
Verifying Image Accuracy
I am now confirming the components and wiring depicted within the generated visual are aligned with the initial prompt. I see an Arduino UNO, two 8BYJ-48 stepper motors, and the standard ULN2003 driver board connected, as expected. I need to make sure the wiring diagram is correct.
Verifying Image Accuracy
I am now confirming the components and wiring depicted within the generated visual are aligned with the initial prompt. I see an Arduino UNO, two 8BYJ-48 stepper motors, and the standard ULN2003 driver board connected, as expected. I need to make sure the wiring diagram is correct.
Podobnie jak wcześniej?
Prompt z postu:
https://www.elektroda.pl/rtvforum/topic4154614.html#21761825
draw graphic guide - ESP32 flashing for firmware change to Tasmota
Rezultaty:
Z postu:
https://www.elektroda.pl/rtvforum/topic4154614.html#21762694
political map of Europe, every country has labelled name and capital city
Rezultaty:
Znacznie lepiej.
Zadanie z obrazkiem.
Change schematic to be hand drawn on paper, do not change component values
Odpowiedzi modelu:
please plug a white phone charger with empty USB A slot into my power strip
Rezultaty:
To samo, ale Thinking Level: High:
Z postu:
https://www.elektroda.pl/rtvforum/topic4154614.html#21762934
change image to remove Flash Memory Chip so only empty pads are visible
Rezultaty:
To samo, ale Thinking Level: High:
Z ciekawości - czy w High zrobi zworkę między 2 a 7 padem?
now make a jumper between pad 2 and 7
Ok, może trzeba precyzyjniej.
now make a jumper between pad 2 and 7 of the removed SOIC chip
Myśli:
Spoiler:
Analyzing the Initial Image
I'm currently focused on the initial macro image, trying to understand all the components in this photo of the PCB. I can see a black printed circuit board, a camera sensor, a blue LED, and several integrated circuits. There's also a finger partially visible. I'm taking this data as a base to start the editing process.
Defining the Edit
I've examined the original image and defined the editing task. My focus is now on precisely identifying the target locations for the jumper wire on the PCB. I'm cross-referencing the SOIC footprint to ensure I'm targeting the correct pads, specifically pad 2 and pad 7, for the jumper connection.
Mapping the Connections
I've determined the exact locations of the required pads: pad 2 is second from the left on the top row, and pad 7 is second from the left on the bottom row of the desoldered SOIC footprint. I am now visualizing the best way for the jumper wire to connect these two points, aiming for realism. I am working on how the wire should arch to bridge the gap.
Visualizing the Wire
I've decided on a fine, insulated wire for the jumper, specifically blue Kynar. I'm focusing on how the jumper will connect pad 2 and pad 7, with a slight arc to give a more realistic appearance. Small solder joints will attach the wire to the pads. I'm making sure it matches the lighting and texture of the original image. The wire shouldn't cover any other components.
Constructing the Prompt
I'm now writing the prompt for editing the image. I have confirmed pad numbering for the desoldered SOIC footprint, re-evaluating the pin numbering for a more accurate description. I'm focusing on providing details about the components, the wire itself, and how it connects, while also making sure the scene lighting and all the other components remain consistent.
Confirming the Pinout
I've re-examined the pin numbering, taking into account both the visual numbering on the PCB and standard SOIC conventions to guarantee accuracy. I have verified that pad 2 is indeed the second from the left on the top row, and pad 7 is the second from the left on the bottom row of the footprint. The goal is to accurately represent a jumper wire between these specific points.
Revising Pin Numbering
I've re-examined the pin numbering system for the SOIC footprint, reviewing both the visual cues and standard conventions to confirm accurate pad identification. This has involved cross-referencing SOIC-8 pinout conventions with the visual clues on the PCB image to ensure that pad 2 and pad 7 are precisely located for jumper wire connection.
I've refined the approach to visualize the jumper wire. The plan is to include realistic details for the jumper wire's connection to the pads.
Refining the Jumper Wire
I'm now zeroing in on the specifics for the jumper wire's appearance. Considering the size of the components, I've decided a 30AWG insulated wire, such as blue Kynar, will provide the appropriate scale and visual detail. I am working on the appropriate size of the solder joints for a realistic result.
Refining the Jumper Wire
I'm now focusing on specific details for the jumper wire and solder joints. Considering existing components' sizes, a fine wire gauge like 30AWG (blue Kynar) should provide the best scale and clarity. I plan to refine the solder joint sizing for realistic results and determine how the arc will be visually.
Analyzing the Initial Image
I'm currently focused on the initial macro image, trying to understand all the components in this photo of the PCB. I can see a black printed circuit board, a camera sensor, a blue LED, and several integrated circuits. There's also a finger partially visible. I'm taking this data as a base to start the editing process.
Defining the Edit
I've examined the original image and defined the editing task. My focus is now on precisely identifying the target locations for the jumper wire on the PCB. I'm cross-referencing the SOIC footprint to ensure I'm targeting the correct pads, specifically pad 2 and pad 7, for the jumper connection.
Mapping the Connections
I've determined the exact locations of the required pads: pad 2 is second from the left on the top row, and pad 7 is second from the left on the bottom row of the desoldered SOIC footprint. I am now visualizing the best way for the jumper wire to connect these two points, aiming for realism. I am working on how the wire should arch to bridge the gap.
Visualizing the Wire
I've decided on a fine, insulated wire for the jumper, specifically blue Kynar. I'm focusing on how the jumper will connect pad 2 and pad 7, with a slight arc to give a more realistic appearance. Small solder joints will attach the wire to the pads. I'm making sure it matches the lighting and texture of the original image. The wire shouldn't cover any other components.
Constructing the Prompt
I'm now writing the prompt for editing the image. I have confirmed pad numbering for the desoldered SOIC footprint, re-evaluating the pin numbering for a more accurate description. I'm focusing on providing details about the components, the wire itself, and how it connects, while also making sure the scene lighting and all the other components remain consistent.
Confirming the Pinout
I've re-examined the pin numbering, taking into account both the visual numbering on the PCB and standard SOIC conventions to guarantee accuracy. I have verified that pad 2 is indeed the second from the left on the top row, and pad 7 is the second from the left on the bottom row of the footprint. The goal is to accurately represent a jumper wire between these specific points.
Revising Pin Numbering
I've re-examined the pin numbering system for the SOIC footprint, reviewing both the visual cues and standard conventions to confirm accurate pad identification. This has involved cross-referencing SOIC-8 pinout conventions with the visual clues on the PCB image to ensure that pad 2 and pad 7 are precisely located for jumper wire connection.
I've refined the approach to visualize the jumper wire. The plan is to include realistic details for the jumper wire's connection to the pads.
Refining the Jumper Wire
I'm now zeroing in on the specifics for the jumper wire's appearance. Considering the size of the components, I've decided a 30AWG insulated wire, such as blue Kynar, will provide the appropriate scale and visual detail. I am working on the appropriate size of the solder joints for a realistic result.
Refining the Jumper Wire
I'm now focusing on specific details for the jumper wire and solder joints. Considering existing components' sizes, a fine wire gauge like 30AWG (blue Kynar) should provide the best scale and clarity. I plan to refine the solder joint sizing for realistic results and determine how the arc will be visually.
No i nie wyszło:
Prompt z obrazkiem:
change hdd to 2.5" SSD Samsung EVO 4TB
Odpowiedzi:
Śladowo lepiej.
Prompt z obrazkiem:
Create a realistic photo showing how would this device look in real life
Odpowiedzi:
Prompt z obrazkiem:
Please analyze the attached photo of the electronic schematic and redraw it as a clean, CAD-style 2D schematic (similar to Eagle or KiCad).
Produce a flat 2D drawing with no perspective and no distortion.
Make all lines strictly horizontal/vertical, organize the layout, and keep the logic identical to the original.
Odpowiedzi:
Znów obraca transoptor...
Tryb high:
Tu jest nieco lepiej, ale widzę co najmniej jeden błąd - połączyły mu się przecinające ścieżki.
please put a common mode choke in the correct place
Odpowiedzi:
Spoiler:
I have placed a common-mode choke coil in the correct location on the circuit board, replacing the temporary wire jumpers. The component has been integrated with the existing circuitry, and I have added a transformer diagram below it to match the board's silk-screened indicators.
I have placed a common-mode choke coil in the correct location on the circuit board, replacing the temporary wire jumpers. The component has been integrated with the existing circuitry, and I have added a transformer diagram below it to match the board's silk-screened indicators.
Brak odczuwalnych zmian względem poprzedniej wersji.
Please annotate sections of this device:
Be detailed and technical
Rezultaty:
Nie wiem jakim cudem AI uznało, że ta grafika łamie zasady...
Tryb high:
Niestety - dalej nonsens. Podstawowe błędy, nawet mostka prostowniczego nie rozpoznało. Dodatkowo... dorysowało rzeczy, których nie ma na PCB.
Testy z:
Wizyjne modele AI do tłumaczenia not katalogowych - testujemy Nano Banana, ChatGPT-Image i inne
translate to english
Rezultat:
Wygląda w porządku.
Podsumowując, w kontekście wykonanych tu testów, elektroniki, schematów, infografik, czy tam "elektronicznych" przeróbek zdjęć krok do przodu jest mały i tylko sporadycznie zauważalny. Każdy może samodzielnie porównać sobie rezultaty z poprzedniego tematu z tym, co pokazałem dzisiaj. Osobiście mogę wyróżnić tylko:
- nieco lepsze schematy prostych rzeczy (z Arduino, włącznik schodowy)
- nieco precyzyjniejsze rozumienie modyfikacji zdjęć elektroniki
- nieco lepsze przerysowywanie schematów (ze zdjęcia kartki do ładnej grafiki)
- również nieco większą wiedzę modelu - np. generuje lepsze mapy...
- nieco mniej halucynacji i całkowicie bzdurnych wyników
Nie zmienia to jednak faktu, że model dalej nie potrafi narysować sensownego schematu dla typowych konstrukcji DIY i halucynuje przy bardziej złożonych poleceniach.
Czy testowaliście już Nano Banana 2? Czy macie jakieś zadanie dla modelu, które chcielibyście, bym wypróbował?
Temat będę jeszcze edytować i dodam kolejne testy, też być może w odpowiedziach
Fajne? Ranking DIY Pomogłem? Kup mi kawę.
